watcher-specs/specs/ocata/implemented/parallel-applier.rst

Make Applier parallel

https://blueprints.launchpad.net/watcher/+spec/parallel-applier

Problem description

Watcher Applier starts Action plan with set of Actions, which are executed in a sequence. After planner-storage-action-plan is implemented, we have new parents field that contains list of action's uuid child action is connected to. Having parents field Watcher can build Directed Acyclic Graph where each independent Action can be run in parallel while actions with some dependencies should fulfill them to be executed.

Use Cases

As an administrator, I would like to be able to run Action plan that will launch actions for every node (compute, network, storage) in parallel.

Proposed change

We can launch several concurrent actions in accordance with their parents.

Workflow engine of Watcher Applier will still use watcher_flow flow, but edges between nodes should be established in accordance with action parents. Unlinked actions with no parents should be connected directly to the watcher_flow and executed firstly.

Main class to modify is DefaultWorkFlowEngine, that defines workflow of actions executing. Since Watcher Default Planner adds actions to DB in accordance with weights (i.e. all migrate actions will be added firstly, then resize actions and so on), Applier will get ordered by weight actions that can be added to watcher_flow flow and linked in one for loop. In this case algorithm of building directed acyclic graph should have O(n) complexity.

There is estimated pseudocode to show part of default planner's workflow:

flow = gf.Flow("watcher_flow")
for a in actions:
    task = TaskFlowActionContainer(a, self)
    flow.add(task)
    if a.parents:
        for parent_id in action.parents:
            parent_action = get_action(actions, parent_id)
            flow.link(parent_action, action, decider=self.decider)

e = engines.load(flow)
e.run()

Alternatives

None

Data model impact

None

REST API impact

None

Security impact

None

Notifications impact

None

Other end user impact

None

Performance Impact

Since it is perfomance related patch, it is expected to have perfomance improvements because of parallel executing of actions instead of sequence. In order to reduce the impact on system performance and stability, it is better to set the maximum number of concurrent.

Other deployer impact

None

Developer impact

None

Implementation

Assignee(s)

Primary assignee: Alexander Chadin <a.chadin@servionica.ru>

Work Items

• Modify watcher/applier/workflow_engine/default.py to build new parallel workflow engine.
• Update the documentation.
• Add appropriate unit tests.

Dependencies

https://blueprints.launchpad.net/watcher/+spec/planner-storage-action-plan

Testing

• Unit tests will be added to validate these modifications.

Documentation Impact

Update the Applier documentation in accordance with new changes.

References

History

None