Both the Boot Orchestration Service (BOS) and Configuration Framework Service (CFS) have issues where race conditions can lead to unexpected or invalid behavior. These issues come in two varieties:
The API servers for both BOS and CFS are multi-process and multi-threaded, allowing multiple requests to be handled simultaneously. In the case where these requests impact overlapping resources, problems can arise. This happens because the API servers are implemented without taking this parallelism into account. An endpoint which modifies a resource will read that resource from the database, modify the resource, and then write the modified resource back to the database. If a delete request for that same resource is handled in the middle of that process, the delete request will succeed, but the resource will still exist in the database.
In general, administrators should avoid making parallel requests that change or delete the same resources.
This issue exists in all versions of BOS. This issue exists in all versions of CFS prior to CSM 1.7.1.
CFS includes operators that run separately from the API server: the CFS batcher, the CFS hardware synchronization agent, and the CFS operator. These operators generally work in a loop where they make a query to the main service API, do some work based on the contents of the response, and possibly send a request back to the main service to change or delete some resources. This offers a window where a resource may change on the API server in the meantime, without the operator being aware of it. This can lead to the operator overwriting changes or taking incorrect actions.
None.
This issue exists in all versions of CFS.