Reference benchmark runs multiple Gatling simulations.
After a write simulation, we wait for asynchronous jobs and refresh the elasticsearch index. The goal is to take in account the async duration and to prevent impacting performance of the next simulation.
But since 9.10 the audit log writer is not anymore handled by asynchronous work (NXP-22109) and the stream processor that writes the audit logs might impact following simulation.
This must be fixed to improve benchmark accuracy.