在自动缩放环境中使用 pm2

Question

I am planning to use the AWS EC2 Container Service to host an auto-scaling group of Node.js + Express instances that expose a REST API. I've seen multiple articles telling me that I should be using pm2 over forever.js in order to ensure that my application restarts if it crashes, that I can have smooth application reloads, etc.

However, I'm a bit confused as to what configuration I should use with pm2 inside of the container. As these instances will be scaled automatically, should I still be running the process manager in "cluster mode"? I want to be sure that I am getting the most out of my instances, and I can't seem to find any definitive answers about whether clustering is necessary in an auto-scaling environment like this (just that pm2 comes with a load-balancer and scaling technique itself).

Answer 1

I would use systemd over pm2 in any case, as its native on most Linux distros now and effectively one less step (with pm2 you still need to make the pm2 daemon a service).

As for running cluster, etc, I think that depends a great deal on what your Node app is doing. As such,I'd probably deploy containers that don't use it, scale as a container rather than inside, and profile for a while. This keeps things inside each container as simple as possible and lets the EVS service manager do its job.

When most folks use the cluster module, they make one worker or maybe two per CPU core. Given that a container is sharing CPU cores with any other containers on the host, it seems like your not getting much bang for the additional complexity.

Answer 2

We have the same situation with AWS EC2 cluster. We created 1 load balancer and 2 server with many CPUs and memory to manage all our applications. Every node.js application has own container and minimum required memory (for example 1GB).

Inside every container we have PM2 with memory limit to restart every process (prevent memory leak) and no limit on CPU or memory for container. Every application has minimum 2 instances inside container (in common 4 instances for both servers).

Also i wrote a small PM2 plugin to automatically scale application inside containers depends on the load and it helps us to scale application to up to MAX CPUs-1. So, you can try to use it https://www.npmjs.com/package/pm2-autoscale and share feedback.

And we have autoscale configuration inside AWS cluster if cluster does not have enough power.