The queuing means you can do with more web workers
Additionally your median times look pretty high, can you get this on a faster CPU?
The queuing means you can do with more web workers
Additionally your median times look pretty high, can you get this on a faster CPU?
Unfortunately, even if the reason for the slowness is this plugin, I must say that this plugin is one of the main plugins of our community and can not be removed from us in any way, and users become heavily depended on it! ![]()
We currently use 24 web workers. Our Discourse is installed on a VM with 24 CPU cores with 2 GHz processing power and 24 GB of RAM. Unfortunately, it is not possible to change the main server now, unless we have no other way to improve performance.
Do you think that in the current situation, increasing the number of web workers can help to improve performance?
More web workers may reduce the queueing
You are trying to eliminate the orange lines from the active web request graph
Performance though will not get any better for your typical median request
Thanks, I understand now. So by looking at the graphs we can see if the number of web workers is enough or not enough for the system.
But according to you, to improve the speed and performance, we now need higher processing power and changing the parameters will not help more.
Be sure to test this at the earliest opportunity and inform the result.
Uma postagem foi mesclada em um tópico existente: Telegram Notifications Plugin (discourse-telegram-notifications)
Qual a origem dos 502s? poucos trabalhadores unicorn? Acabamos de mudar de servidor e os usuários têm relatado 502s ocasionais após a mudança. Temos um servidor 6C/16GB e definimos unicorn_workers para 16 (antes de mudarmos de servidor, estava em 12), db_shared_buffer para 4GB, db_work_mem para 60MB.
Ter muitos trabalhadores unicorn poderia causar 502s? (especulando, já que tínhamos 4 a menos antes da migração do servidor)
Estou tentando ver se é a configuração que precisamos ajustar ou algo mais que precisamos resolver.
Obrigado.
Sim, isso parece alto. Defina seus workers do unicorn entre 1 e 2 por núcleo.
Para 6 núcleos, não deve ser mais do que 12. Dado o seu buffer compartilhado bastante grande e meu instinto, eu o definiria para cerca de 10 com tais especificações de servidor.