cloudkitty-specs/specs/train/reset_scope_state.rst
Luka Peschke df4d9cbd10 Spec: Allowing to get/reset the state of a scope
Change-Id: I88bf7453b370f383adc5bb3c7d8a44bb04c6eed2
2019-05-17 11:48:16 +02:00

4.9 KiB

Allowing to get/reset the state of a scope

https://storyboard.openstack.org/#!/story/2005395

Problem Description

With the current state of CloudKitty, if an admin wants to reset calculations for a given scope for any reason (changes in cloudkitty's collection configuration, update of some rating rules...), he has to stop all the processors, update the state of said scope in the database, and restart all the processors. Furthermore, there is no way for an admin to get the current state of a scope, unless he directly queries the SQL database.

Proposed Change

Warning

This spec only covers the global schema. If it is accepted, three related specs detailing each part of the solution will be proposed.

In order to solve this, the following three changes are proposed:

  • Add a delete method to the v2 storage interface allowing to delete all dataframes after a specific timestamp for one or several scopes. This will be the subject of another spec.
  • Add a v2 API endpoint allowing to reset the state of one or several scopes. This will be the subject of another spec.
  • Add a v2 API endpoint allowing to retrieve the state of one or several scopes. This endpoint will be a replacement for v1's /report/tenants. This will be the subject of another spec.

The proposed logic is the following:

  • The admin sets the status of a scope through the API. He can regularly check if the status of the previously updated scope has been updated through the scope state endpoint.
  • The API endpoint sends an AMQP notification indicating that one or several scopes should be reset to a given state. Once an AMQP reply has been received by the API, either a 202 Accepted or 400 Bad Request status code will be returned (it is implied that authentication-related issues leading to 401 or 403 errors will have been handled before reaching this point).
  • Upon receiving a notification, the associated AMQP listener goes through the following steps:
    • The notification is parsed and validated. An AMQP reply indicating if it is valid or not is returned to the API. If the notification was invalid, the listener stops here.
    • For each scope, the following steps are applied:
      • To prevent a scope from being processed while its state is being reset and its data being deleted, the listener acquires a tooz lock. This operation is blocking.
      • Once the lock is acquired, all the data after the given state is deleted from the storage backend.
      • After the data has been deleted, the state of the scope is set to the given state.
      • Once this is done, the lock is released.

Alternatives

This could be done without the AMQP notification. However, this would imply to execute the locking and data update/deletion from the API, which would lead to request timeouts.

Data model impact

None.

REST API impact

The REST API impact will be detailed in the specs related to the API endpoints described above.

Security impact

This will allow admins to delete user data through the API. This must be used with care.

Notifications Impact

This will add a notification and a new AMQP listener to the processor. The exact change will be described in the associated spec.

Other end user impact

None.

Performance Impact

Depending on the amount of data associated to a scope, the deletion may take a consequent amount of time and affect the performance of the database.

Moreover, the scope to reset will be locked during data deletion, and processors will thus not be able to process it.

Other deployer impact

This will help admins, as resetting a scope won't require to manipulate the database directly anymore.

Developer impact

None

Implementation

Assignee(s)

Primary assignee:

peschk_l

Other contributors:

jferrieu

Work Items

  • Write a spec to add a delete method to the v2 storage interface allowing to delete all dataframes after a specific timestamp for one or several scopes.
  • Write a spec to add a v2 API endpoint allowing to reset the state of one or several scopes.
  • Write a spec to add a v2 API endpoint allowing to retrieve the state of one or several scopes. This endpoint will be a replacement for v1's /report/tenants.

Dependencies

None.

Testing

Testing will be detailed in each of the associated specs.

Documentation Impact

  • The API reference will be updated
  • Some documentation and examples on how to reset the state of a scope will be added.

References

None