We are trying to monitor response times of HTTP requests using stream using the time_taken value. When we enable HTTP persistent connections (aka http keep-alive, connection reuse or oneconnect in our f5 load balancer) and we run a load test, we notice that the time_taken reported by stream increases significantly and incorrectly. We can tell via other monitoring tools and packet captures that the time_taken is incorrect. Since the tcp connections get reused, it seems like it is mixing up which request belongs to which response and reporting longer times. Is this a known issue or are we missing something in our config?