[SR-Users] http sync / async final tests

Just to share, and thanks for your valuable help. My conclusion after tests (on my scenario) is http backend: is a golang http server with a 250ms delay to answer, and answering back a json, running in 10 replicas inside a K8 with 4 nodes. test (from kamailio perspective): receive INVITE send 100 call the http, get the result, get a key from the result and set it as the value of Contact header send 302 receive ACK (discard on sync scenario because it’s stateless, use it in async to end transaction) Kamailio host: 1vcpu kamailio 3.8.4 bookworm running inside containerd (with podman) 16GB RAM sipp: run from the K8 nodes hosts (directly from OS) (everything is LAN connected on 2 different physical servers, hypervisor is VMWare i guess 7.0.3) using http_client (sync with stateless): 8 children -M8 -m64 no more than 60 calls/sec (15cps on each sipp instance) until sipp start to retransmit the INVITES using async_http_client: 8 children -M8 -m2048 it’s been ~36 hours now at 250cps each sipp instance with 0 errors 0 retransmission and the kamailio host is userlab@sblb1:~$ vmstat 1 procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu----- r b swpd free buff cache si so bi bo in cs us sy id wa st 4 0 0 10285460 364684 5444240 0 0 0 22 36 68 0 0 100 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5284 6822 27 14 58 0 0 4 0 0 10285460 364684 5444240 0 0 0 0 5306 6752 29 15 57 0 0 3 0 0 10285460 364684 5444240 0 0 0 0 5304 6823 26 16 57 0 0 2 0 0 10285460 364684 5444240 0 0 0 12 5287 6761 26 16 57 0 0 2 0 0 10285460 364684 5444240 0 0 0 0 5337 6838 27 14 58 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5101 6724 25 17 59 0 0 3 0 0 10285460 364684 5444240 0 0 0 0 5066 6650 26 14 60 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5306 6829 23 17 59 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5219 6725 23 19 58 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5274 6770 26 13 61 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5295 6756 26 18 56 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5235 6786 22 16 62 0 0 1 0 0 10285460 364684 5444240 0 0 0 0 5315 6833 21 20 59 0 0 at least for my scenario, it worths to pay the price of the memory tradeoff. again, Ben, Alex, Sergio, all, thanks for your help. regards