Session reset or instance restart to clear stale pull-request approvals?

In Bitbucket Data Center 8.5, a pull request can continue to display an approval from a user who has been removed from the reviewers group. The goal is to make the approval list reflect current group membership immediately, without waiting for the cached entry to expire.

The reviewer-approval cache is keyed to group membership at request time, and no configurable TTL or invalidation API is documented. The available workarounds are a user session reset or a full instance restart, but Data Center may use separate cache layers, and behavior differs from Bitbucket Server. Version 8.0 introduced a new session store, so earlier versions may not exhibit the same pattern.

Which approach is safe for a production cluster: session reset or node restart? Does session reset invalidate the cache across all cluster nodes or only for the affected session? Is there a documented cache-invalidation endpoint for the reviewer-approval cache in Data Center 8.5?