Due to external circumstances our current implementation of OLM upgrade tests in Claire sometimes fails. In most cases some flakes are happening due to strange timing issues. What causes them will warrant more investigation, but these are almost impossible to reproduce.

What seems to have happened in the latest flaky test run has been most likely caused by the index updates by release engineering. Brief description of situation: In one iteration of the upgrade test, it failed on timeout, then the new iteration of test has started. Previous iteration has finished in the middle of the current run so the operator pod has changed and started the reconciliation process of the brokers but the version of the operator wouldn't fit and and due to the ongoing reconciliation, some of the resources created by operator were deleted at the moment Claire was checking them ...

We need to update/upgrade the test itself in a way that previous iterations won't influence subsequent ones.