We are currently using v2.2.24 of the community version of HAProxy. We are using stick tables to test for excessive requests. While we are spreading traffic equally across our HAProxy estate via F5 GLB using pure round robin, I do not see the stick table values reflect this. I understand that the last writer wins in the community version, but if I have an abusing app, should I not see the numbers on all peers converging? I see a few of the peers getting hit and they have similar numbers, but others are nowhere near the counts I expect them to be. Now maybe some of my apps are caching the HAproxy IP, but even so, with peer replication, shouldn’t the stick table rates be similar? Is there a sweet spot for the number of peers before the replication between them never finishes prior to another peer being updated? I am trying to come up with thresholds for our environment, but cannot truly explain what I am seeing in our logs. Any guidance provided will be most welcome.
Best regards,
Steve
1 post - 1 participant