11–13 of 13 posts
FPM workers get stuck, default settings need to be changed, can’t handle long-running connections well, ends up using a lot of ram. More log places to check when things go wrong.
You end up with more issues.
What killed Apache for me were the complexity of various rules and how challenging it is to define a proxy for a WebSocket connection. NGINX is better but still not great.
For me anything can be connected to anything else easily and quickly. A network connection, a socket, is a pipe containing a full duplex stream. So, just pipe one socket to another, and vice versa, and you have a very low level proxy. The rules to define such for a users customization is just as simple as what should go where.
Knowing all of that, and that HTTP is not full duplex, I serve HTTP and WebSockets from the same port with a preference for WebSockets by default. I can also pipe other streams to sockets for lower CPU overhead. I have not tested if my server is faster and less CPU overhead than NGINX yet, but it does concurrency very well.