Hi Aleks, On Sat, Feb 21, 2026 at 01:54:36AM +0100, Aleksandar Lazic wrote: > Hi dear HAProxy Team. > > I have created an benchmark run which uses 7 servers and now I would like to > test http backends h1.1, h2 and h3. > The timing for haterm as backend server is perfect, thank you. :-) > > I run haterm with that command and I'm able to connect with http/1.1 > plain+tls, h2 and h3, cool.
Thanks for your kind words. Indeed, we were also missing something like this for tls/h2/h3 and whlie I thought we'd have to reimplement something external relying on generic stacks, Fred suggested to instead rely on the haproxy lower layers. Initially I thought it would be very complicated to repurpose it but Fred proved me wrong and that's great :-) We're also going to work on the client now (more complicated since many more options), because right now using h1load piped to haproxy for h2/h3 is not convenient, and h2load returns variations of +/-50% between runs because it's connection-driven, and once a connection has sent/processed all its requests, it's no longer used, so a long part at the end of the test is the cool down period but it participates to the average measure! Also not seeing variations nor anomalies in real time makes it mostly useless for many use cases, yet at the moment we don't have better. Once that's done I'll update the dpbench project to add references to these new tools, hoping others will find them useful as well. Cheers, willy
