0

I'm currently running a MeteorJS application on Amazon. MeteorJS has a plugin that allows each server to act as a load balancer or a worker for said load balancers. I can spin up any number of workers with no problems as the workers check in with the load balancers and they are added to the pool.

The issue I'm having is scaling the load balancers. Each load balancer needs a valid DNS A record. I was thinking that perhaps I would add a bunch of A records pointing to IP addresses that will be used if I need to spin up additional load balancers. However, my concern is that this will result in poor performance due to users browsers trying multiple addresses.

Is it poor practice to have dead A records behind your primary domain? i.e.

somewebsite.com 
loadbalancer1.somewebsite.com -> Online
loadbalancer2.somewebsite.com -> Online
loadbalancer3.somewebsite.com -> Offline
loadbalancer4.somewebsite.com -> Offline

I was hoping that Amazon Route 53 would help me with this, but they only provide failover services. Ideally these records that point to offline servers would not be published

1
  • 2
    AFAIK - If you attach a health check to the dns record in route53 and that endpoint is down it will not publish the record. The only way to see that it exists is using the api call or control panel as the account owner
    – Drifter104
    Apr 11, 2016 at 16:51

1 Answer 1

1

Ideally these records that point to offline servers would not be published

Ideally this is the worst case because it means changing published DNS servers, which is only possible with a short expiration which is bad.

The normal use case is to have a SMALL number of PROPER load balancers behind ONE or a small number of IP addresses (which is something proper load balancers can handle).

They then distribute the work.

2
  • Thanks for the answer @TomTom. My only reason for not wanting to run multiple load balancers all the time is due to cost. Currently my web app has very few users, but I am trying to plan out how I will scale it in the future. The impression I'm getting is that I need to bite the bullet and maintain a few very stable load balancers (possibly not MeteorJS ones, something like haproxy or nginx). Apr 11, 2016 at 16:21
  • Well, load balancers are QUITE "primitive" - as in not resource intensive. You will find that if you use a proper one (ngix is not bad) then you can get away with the usual 3-4 that you need for high availability (double redundancy - we talk of possibly quite small virtual machines, but you NEED a minimum of 3 for decent high availability)) for a LOT.
    – TomTom
    Apr 11, 2016 at 16:48

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .