Watcher Decision Engine has an external cluster data model (CDM) plugin interface which gives anyone the ability to integrate an external cluster data model collector (CDMC) in order to extend the initial set of cluster data model collectors Watcher provides.
This section gives some guidelines on how to implement and integrate custom cluster data model collectors within Watcher.
In order to create a new model, you have to:
First of all, you have to extend the BaseClusterDataModelCollector base class which defines the execute() abstract method you will have to implement. This method is responsible for building an entire cluster data model.
Here is an example showing how you can write a plugin called DummyClusterDataModelCollector:
# Filepath = <PROJECT_DIR>/thirdparty/dummy.py
# Import path = thirdparty.dummy
from watcher.decision_engine.model import model_root
from watcher.decision_engine.model.collector import base
class DummyClusterDataModelCollector(base.BaseClusterDataModelCollector):
def execute(self):
model = model_root.ModelRoot()
# Do something here...
return model
@property
def notification_endpoints(self):
return []
This implementation is the most basic one. So in order to get a better understanding on how to implement a more advanced cluster data model collector, have a look at the NovaClusterDataModelCollector class.
At this point, you have a fully functional cluster data model collector. By default, cluster data model collectors define a period option (see get_config_opts()) that corresponds to the interval of time between each synchronization of the in-memory model.
However, in more complex implementation, you may want to define some configuration options so one can tune the cluster data model collector to your needs. To do so, you can implement the get_config_opts() class method as followed:
from oslo_config import cfg
from watcher.decision_engine.model import model_root
from watcher.decision_engine.model.collector import base
class DummyClusterDataModelCollector(base.BaseClusterDataModelCollector):
def execute(self):
model = model_root.ModelRoot()
# Do something here...
return model
@property
def notification_endpoints(self):
return []
@classmethod
def get_config_opts(cls):
return super(
DummyClusterDataModelCollector, cls).get_config_opts() + [
cfg.StrOpt('test_opt', help="Demo Option.", default=0),
# Some more options ...
]
The configuration options defined within this class method will be included within the global watcher.conf configuration file under a section named by convention: {namespace}.{plugin_name} (see section Register a new entry point). The namespace for CDMC plugins is watcher_cluster_data_model_collectors, so in our case, the watcher.conf configuration would have to be modified as followed:
[watcher_cluster_data_model_collectors.dummy]
# Option used for testing.
test_opt = test_value
Then, the configuration options you define within this method will then be injected in each instantiated object via the config parameter of the __init__() method.
Here below is the abstract BaseClusterDataModelCollector class that every single cluster data model collector should implement:
Build a cluster data model
Synchronize the cluster data model
Whenever called this synchronization will perform a drop-in replacement with the existing cluster data model
In order for the Watcher Decision Engine to load your new cluster data model collector, the latter must be registered as a named entry point under the watcher_cluster_data_model_collectors entry point namespace of your setup.py file. If you are using pbr, this entry point should be placed in your setup.cfg file.
The name you give to your entry point has to be unique.
Here below is how to register DummyClusterDataModelCollector using pbr:
[entry_points]
watcher_cluster_data_model_collectors =
dummy = thirdparty.dummy:DummyClusterDataModelCollector
At this point, you have a fully functional cluster data model collector. However, this CDMC is only refreshed periodically via a background scheduler. As you may sometimes execute a strategy with a stale CDM due to a high activity on your infrastructure, you can define some notification endpoints that will be responsible for incrementally updating the CDM based on notifications emitted by other services such as Nova. To do so, you can implement and register a new DummyEndpoint notification endpoint regarding a dummy event as shown below:
from watcher.decision_engine.model import model_root
from watcher.decision_engine.model.collector import base
class DummyNotification(base.NotificationEndpoint):
@property
def filter_rule(self):
return filtering.NotificationFilter(
publisher_id=r'.*',
event_type=r'^dummy$',
)
def info(self, ctxt, publisher_id, event_type, payload, metadata):
# Do some CDM modifications here...
pass
class DummyClusterDataModelCollector(base.BaseClusterDataModelCollector):
def execute(self):
model = model_root.ModelRoot()
# Do something here...
return model
@property
def notification_endpoints(self):
return [DummyNotification(self)]
Note that if the event you are trying to listen to is published by a new service, you may have to also add a new topic Watcher will have to subscribe to in the notification_topics option of the [watcher_decision_engine] section.
The Watcher Decision Engine service will automatically discover any installed plugins when it is restarted. If a Python package containing a custom plugin is installed within the same environment as Watcher, Watcher will automatically make that plugin available for use.
At this point, you can use your new cluster data model plugin in your strategy plugin by using the collector_manager property as followed:
# [...]
dummy_collector = self.collector_manager.get_cluster_model_collector(
"dummy") # "dummy" is the name of the entry point we declared earlier
dummy_model = collector.get_latest_cluster_data_model()
# Do some stuff with this model