In ephemeral, we see a huge performance downgrade of HBI endpoints when the Kessel Phase 1 is enabled. The groups endpoints are the ones that are least impacted here, with GET /groups having roughly the same performance with Kessel as with legacy RBAC. The other endpoints range from 233% to 2517% performance decrease. These results are calculated with the default 10s quantization interval and the quantization staleness set to 1 (100%). If I lower the quantization interval, the performance is even worse. I think the performance in ephemeral will need to be increased in order to make our test run times usable for everyday testing. Some of the most used endpoints by our tests are:

• GET /hosts/<host_ids> : 522% performance degradation
• DELETE /hosts/<host_ids>: 701% performance degradation
• GET /host_exists: 1900% performance degradation

If these keep being so slow, I don't think we can speed up the tests enough by other ways. Here is the summary of the results in ephemeral: https://docs.google.com/spreadsheets/d/1ZZcD14aYgIMs9o70xbJjeCC64w7NkS6kppS14_tNZo4/edit?gid=806725582#gid=806725582

Also, if you are interested in full results with a lot more details (including difference between single-resource and multi-resources (using bulkcheck) requests), please take a look at the comments in this Jira: https://issues.redhat.com/browse/RHINENG-23896

Related slack thread: https://redhat-internal.slack.com/archives/C08GTNAFZ55/p1771338214616809