This message was deleted.

I've used haproxy in tcp passthrough mode (no ssl termination or proxy) in the past. IIRC we did leastconn to the backend servers. Connections are long lived so tendency of traffic clump after server side restarts. further haproxy notes. • Send logs to /dev/log instead of that local0 nonsense. You'll need to add the socket via this in rsyslog ``$AddUnixListenSocket /var/lib/haproxy/dev/log`` • enable seamless reloads. watchdog of 3-600s should be fine. • haproxy::listen won't work. Use haproxy::frontend and haproxy::backend

I think the main thing you are missing not using the loadbalancer is the balancing part, i believe the agent progresses through the list of broker uris until one responds

we could shuffle the URI list for each node, but I also think the loadbalancer is the better approach

its a good option if you dont have a loadbalancer, its flexibility. But in practice the response and timeout to progressing over a list would make the system less responsive

did you ever measure how much load the connections create? is loadbalancing even required? if we add URIs to two compilers and compiler 1 handles all connections until it goes down for maintenance, that would be nice as well

I dont think i have anything documenting the load per connnection

while I'm a fan of haproxy/nginx, I hate f5 hardware, f5 as the company and the f5 module so much and many "enterprise" customers use it, and I would like to avoid it 😄

Need to connect you to with my F5 reps, they are lovely people 🙂