Server load by secondary tasks


#1

Do we have any metric anywhere about the server load for any tasks other than methods or publications? Like, starting a connection, deliver content to client, keep up heartbeat, maybe some other actions performed by the server for each client?

Even though it might be very tiny percentage, multiplying it to thousands of users can be a substantial number.
Since my load test considers only methods and publications that are called from another server-app, and therefore no client is involved during such testing (at least web-browser related resources are not sent to the connected server-app), it would be great to know how correct my numbers/expectations are.

I understand that the answer could be ‘it depends’, but maybe someone could explain a bit in details about the exact steps/tasks it takes for the server to serve each client?


#2

I forgot to mention the main thing - reactive computations, as if there are for example 10 opened browser windows on the same page and there is a helper changing every second on that page, it would need to be served to all the clients who are on this page, right? and this invalidation means minimongo updating a value (or complete subscription?) every second…


#3

Here’s a thread about testing your Meteor app using PhantomJS with your own AWS EC2 instances to test with distributed, headless browser instances that will actually download and test your app using a WebDriver test case. It makes use of a good cloud load tester called www.redline13.com:

https://forums.meteor.com/t/poor-galaxy-meteor-performance-serving-small-bursts-of-users-load-test/38671