I see that we can specify the max connection counts for each instance of an app — is there a limit to this? Can we set any connection count that the CPU/MEM can handle or is there an implicit limit in the networking / routing stack?
I’m deploying an app that accepts websocket connections, I’m hoping I measure and configure each instance to accept as many connections as it can handle without worrying about bottlenecks in the load balancer.
You can set them pretty high. There are no fixed limits on the network infrastructure, but I’m sure there are bottlenecks we haven’t discovered yet. These will manifest in different ways, depending on where websocket connections originate.
For what you’re doing, I’d run as many smaller VMs as you can and let it spread those connections around as much as possible.
We have people running micro-1x nginx VMs that handle ~2k concurrent connections.
For what you’re doing, I’d run as many smaller VMs as you can and let it spread those connections around as much as possible.
Sounds good, thanks. That’s what I needed to know. And yeah, I’m assuming an average case of 1000 connections on a micro-1x.
Oh that should work great. Hopefully you’re using a runtime that doesn’t use much RAM per connection though. 
Yeah, most likely Go. Think each connection is on the order of ~4KB to 10KB, so should be possible to fit between 1000 and 10,000 connections in a micro 1X.