Used Zammad version: 3.6.0-1605614796.0157dc25.buster
Used Zammad installation source: Debian package
Operating system: Debian 10
Browser + version: Various
Expected behavior:
Zammad runs smoothly and shows changes within 2-3 seconds. The websocket server should not use much CPU as it just forwards messages (?)
Actual behavior:
Zammad runs extremely slow (mean delay around 30 seconds, with peaks over 1 minute). The websocket server runs at 100% CPU for most of the time.
Steps to reproduce the behavior:
Standard installation with chat integration enabled
The problem starts when the number of supporters goes above around 10-15. In the peak we have 40-50 supporters concurrently online with support coming in just via chat, and then the system is rather unusable.
We have web workers set to 8 and schedulers set to 4 on a 16-core machine. The CPU utilization of all these processes is fine.
So updating Zammad should solve the issue.
If it does not, remove your agents from their organisation as this may be a further factor that decreases systems performance.
Also web workers are not relevant to this topic directly, because it’s the websocket server being under heavy load (that’s one single process). This only is an issue if you have a lot concurrent agents and big organizations.