Server-Side Entitlement Status

Once upon a time the client carried a complete implementation of code for determining entitlement status (red/yellow/green) to accommodate the disconnected use case. Unfortunately, because this was eventually also needed server side for healing/auto-attach and management tools, this logic had to be duplicated in the candlepin server as well. This hampered efforts to roll out new entitlement features because all work had to be done twice, kept in sync, and the client would need to be updated before the feature was usable.

However, we will be transitioning to a model where the server is responsible for determining this status and the client will request it on the fly. This allows the logic to exist in only one place and language, and enables us to change it in a centralized fashion by publishing new JavaScript rules in Candlepin. Once uploaded in hosted, these rules will begin to trickle out to downstream Candlepin servers in SAM/SE through the manifests.

The upload in hosted is as before, IT basically can either roll out a new Candlepin RPM containing new rules, or use an API call to update them on the fly.

For disconnected systems, the client will report that entitlement status is effectively “unknown”. We will likely indicate whether we have access to the content for an installed product however. (pending mockups) The only compliance logic remaining in subscription manager will be this check if a product has a certificate that grants access to its content, and some rudimentary date comparisons in some situations when we need to display a certificate is expired or about to expire. All calculations of socket/ram coverage and red/yellow/green will defer to the server.

For a registered system, all behaviour should be exactly as before, only now the client is requesting status from the server on the fly rather than calculating it locally.

Detailed below there will be a caching mechanism we can fall back on in the event the server is temporarily unreachable.

Tasks

  1. Add GET /consumers/{uuid}/compliance to python-rhsm connection class.
  2. Add method to get list of installed product IDs to product directory if not there already.
    1. This should replace some uses of the cert sorter which will now be a little slower (as they require server communication).
  3. Encapsulate this within existing CertSorter Python class.
    1. Will have to pass in a connection object, these are created in relatively few places. When updating constructors, look for any uses which are just getting a list of installed IDs and switch to using the product directory method mentioned above instead.
    2. When creating/reloading, we will request status from the server and expose the exact same information via the existing Python object. This will impact performance some but the reload should only be called when entitlement status has changed.
  4. When system is not registered:
    1. Modify cert sorter to indicate this.
    2. Update uses of cert sorter to show no indication of red/green/yellow.
    3. Display prominently in UI and CLI that the system is not registered and do not show entitlement status. (pending mockups from mreid)
      • May also need a way to show that a product is technically covered in that we can access its content, but we can’t use red/green/yellow as this will be confusing to users.
  5. Eliminate all old entitlement status calculation code in client, and all tests for it.
    1. Add some new cert sorter tests with mocked status from the server. A small test utility class to simulate the server JSON will be useful for this.
    2. Any other tests that are leveraging cert sorter will probably need updating. Utility may be useful for this. This may replace some mocked entitlement cert code.
  6. When a system is registered, but unable to reach the server:
    1. Whenever we fetch status from server, write a cache to disk.
      • Because this is just a full file write and not an append / modification, we should not need to worry about concurrent writes between say the GUI and the daemon running in the background. Last write will win, which is fine for our purposes, so no file locking is necessary.
    2. If we are unable to reach server but registered, display this prominently in the UI and indicate that we are unable to reach server temporarily. (pending mockups from mreid)
    3. If the cached compliantUntil time is greater than the current time, use the cached data to display product status normally instead of fetching from server.
    4. If the cached compliantUntil time is less than the current time, fall back to showing no status for products as per the disconnected use case, as we know we’re showing inaccurate data.
Last modified on 26 April 2024