RuntimeConfigBundle

This bundle provides a way to inject parameters into services at runtime by exposing a RuntimeParameterBag service, which functions exactly like Symfony2's own ParameterBags.

As-is, Symfony2's service container is compiled and cached to disk, which makes it difficult to inject dynamic parameters. By exposing a ParameterBag service, we can inject values returned from its get() method into other services.

One reason you might want support for dynamic parameters would be to implement feature flags/flippers, as are used by GitHub and Flickr. More info on the history behind this bundle may be found on the symfony-devs mailing list.

Installation

Submodule Creation

Add RuntimeConfigBundle to your vendor/ directory:

$ git submodule add https://github.com/opensky/OpenSkyRuntimeConfigBundle.git vendor/bundles/OpenSky/Bundle/RuntimeConfigBundle

Class Autoloading

Register the "OpenSky" namespace prefix in your project's autoload.php:

# app/autoload.php

$loader->registerNamespaces(array(
    'OpenSky' => __DIR__'/../vendor/bundles',
));

Application Kernel

Add RuntimeConfigBundle to the registerBundles() method of your application kernel:

public function registerBundles()
{
    return array(
        new OpenSky\Bundle\RuntimeConfigBundle\OpenSkyRuntimeConfigBundle(),
    );
}

Configuration

RuntimeConfigBundle Extension

The RuntimeParameterBag may be configured with the following:

# app/config/config.yml

opensky_runtime_config:
    provider: parameter.provider.service
    cascade:  true
    logging:
        enabled: true
        level:   debug

These settings are explained below:

• provider: A service implementing ParameterProviderInterface. If you are using Doctrine ORM as your datasource, this could be an EntityRepository.
• cascade: If true, calls to get() will cascade to the service container when the parameter is undefined in the runtime configuration. This will not change the behavior of has() or all(), which always only consider parameters from the runtime configuration provider.
• logging.enabled: Whether to enable logging access to undefined parameters, regardless of whether service container cascading is enabled. If you are using Monolog, logs will be sent to the "opensky.runtime_config" channel.
• logging.level: Log level to use (should be a LoggerInterface method).

Note: when using cascade, it's a good idea to define default values for your runtime configuration parameters in your service container. This will help avoid an undesirable ParameterNotFoundException if you happen to fetch a parameter that is not yet defined in your runtime configuration.

Injecting Parameters

Consider the scenario where "my.service" depends on a dynamic parameter "my.service.enabled".

Runtime parameters may be conveniently injected by abusing the anonymous service syntax in XML configurations:

# Resources/config/my_service.xml

<service id="my.service" class="MyService">
    <argument type="service">
        <service class="stdClass" factory-service="opensky.runtime_config" factory-method="get">
            <argument>my.service.enabled</argument>
        </service>
    </argument>
</service>

Unfortunately, the YAML format does not yet support defining anonymous services. Parameter injection is still possible, but more verbose:

# MyBundle/Resources/config/my_service.yml

my.service:
    class: MyService
    arguments:
        - @my.service.enabled

my.service.enabled:
    public: false
    class: stdClass
    factory_service: opensky.runtime_config
    factory_method: get
    arguments:
        - my.service.enabled

Note: in both cases (anonymous and labeled services), Symfony2 requires that we define the class on our service definition. The above examples use "stdClass" as an arbitrary placeholder to satisfy CheckDefinitionValidityPass. In reality, our service is simply a means to lazily load our parameter. The value returned by get() can be anything (e.g. object, scalar, array).

Cascade Mode

If you have enabled cascade mode, get() will attempt to fetch undefined runtime parameters from the service container before throwing an exception.

Building upon the previous XML example, this would look like:

# Resources/config/my_service.xml

<parameters>
    <parameter key="my.service.enabled">false</parameter>
</parameters>

<services>
    <service id="my.service" class="MyService">
        <argument type="service">
            <service class="stdClass" factory-service="opensky.runtime_config" factory-method="get">
                <argument>my.service.enabled</argument>
            </service>
        </argument>
    </service>
</service>

In this example, get('my.services.enabled') would return false even if the parameter was not defined in the runtime configuration. This is a safe way to introduce new parameters, which might not yet be available from your provider at the time of deployment.

Note: parameters sourced from the runtime configuration provider are not resolved for placeholder syntax (i.e. "%reference%"), unlike those defined in the service container.

Recipe: Interpreting Parameter Values as YAML

If you're using Doctrine ORM (or any database) to hold your parameters, you will likely implement a CRUD interface to define and edit parameters via an admin controller in your application.

Additionally, this allows us to add custom behavior to our ParameterProvider. For instance, we can use Symfony2's YAML component to interpret parameter values stored in the database as strings.

Consider the following Entity:

# MyBundle/Entity/Parameter.php

namespace MyBundle\Entity\Parameter;

use Doctrine\ORM\Mapping as ORM;
use OpenSky\Bundle\RuntimeConfigBundle\Entity\Parameter as BaseParameter;
use Symfony\Component\Validator\Constraints as Assert;
use Symfony\Component\Validator\ExecutionContext;
use Symfony\Component\Yaml\Inline;
use Symfony\Component\Yaml\ParserException;

/**
 * @ORM\Entity(repositoryClass="MyBundle\Entity\ParameterRepository")
 * @ORM\Table(
 *     name="parameters",
 *     uniqueConstraints={
 *         @ORM\UniqueConstraint(name="name_unique", columns={"name"})
 *     }
 * )
 * @Assert\Callback(methods={"validateValueAsYaml"})
 */
class Parameter extends BaseParameter
{
    /**
     * @ORM\Id
     * @ORM\Column(type="integer")
     * @ORM\GeneratedValue
     */
    protected $id;

    public function getId()
    {
        return $this->id;
    }

    public function validateValueAsYaml(ExecutionContext $context)
    {
        try {
            Inline::load($this->value);
        } catch (ParserException $e) {
            $context->setPropertyPath($context->getPropertyPath() . '.value');
            $context->addViolation('This value is not valid YAML syntax', array(), $this->value);
        }
    }
}

Several things are happening here:

• We must map an ID field, as the base Parameter class only defines essential name and value fields.
• The base class defines assertions for name and value fields (in groups, which can be easily disabled); however, the mapped superclass does not define a unique constraint on the name, so that is necessary.
• A callback assertion is used to check that the value property is valid YAML.

The above Entity class is complemented by the following EntityRepository, which serves as the ParameterProvider for the RuntimeParameterBag:

# MyBundle/Entity/ParameterRepository.php

namespace MyBundle\Entity\Parameter;

use OpenSky\Bundle\RuntimeConfigBundle\Entity\ParameterRepository as BaseParameterRepository;
use Symfony\Component\Yaml\Inline;

class ParameterRepository extends BaseParameterRepository
{
    public function getParametersAsKeyValueHash()
    {
        return array_map(
            function($v){ return Inline::load($v); },
            parent::getParametersAsKeyValueHash()
        );
    }
}

The base ParameterRepository already fetches name/value pairs from the database via a DQL query. Using array_map(), we can easily interpret those values through the same YAML component method.

Note: although we validate the Entity, it's possible that a value might have been manually altered in the database and contain invalid YAML when parameters are fetched for provision. If this is a concern, you may want to gracefully handle thrown ParserExceptions within getParametersAsKeyValueHash().