Yes， we deploy it on our node
So when this happens check your signaling node, and the prosody process CPU usage? How does that look like?
There are several threads about optimizing that and using epoll as by default it comes with select as a network backend …
In the node, only 30% or more CPU are using.
I think maybe it’s not the problem.
Did you check prosody process specifically?
how do you test 120 users+ ? with torture ?
With multiple servers running selenium, pointing the same room.
I’ve tested 80 per server with one of the strongest machines on AWS.
That makes 80 x #servers.
So you get the number of clients you want to test.
tks, i also do it but weston?
We had the test it by the real people, not by the selenium.
But It’s not a good way.
did you fix your issue ?