There are things to consider:
Takes a lot of time to process with the current CI architecture. This adds 12 CI jobs, each runs for 1 hour. There are three dedicated machines, which can run stress tests (one machine per Linux amd64, Linux arm64, and FreeBSD amd64), that's additional 4 hours of load per machine per branch, in total 8 hours per machine. Currently each of these three machines dedicated to stress tests is busy for 12 hours a day (6 pipelines with 2 jobs each). With this MR adding 8 hours that's 20 hours of load per day. Obviously not practical for daily execution as some result would arrive almost a day later. Tho it should work if executed only on Sunday, with result arriving by Monday morning. (Of course, a more scalable solution like AWS is warranted.)
namedprocesses borderline amount of DoH queries.
namedcan process around 21.5 million TCP queries per hour with two generators each sending up to 3000 queries per second (because of
tcp-clients: 50;it's more like 2500 each). With DoH it falls down to 20.4 million for amd64 and 19.9 million for arm64 (and unknown for FreeBSD as it needs isc-private/devops!11 integrated first for Flamethrower acquire DoH capability). With minimum number of TCP queries required to pass set to 19,440,000, this is a close call.
DoT performance is low, very low on FreeBSD. Instead of 21.5 million TCP queries or 20.4 million DoH queries processed, there are only 18.1 million DoT queries processed on amd64, 17.8 million on arm64, and 13.5 million on FreeBSD (where generators send only 2×1300 qps).