In one of my projects we are using Orleans, the virtual actor framework from Microsoft. Your Orleans backend(the cluster) can span out over multiple servers(silos).
In a cluster there are 2 types of communication happening:
- silo-to-silo communication(through the silo port)
- client-to-silo communication(through the gateway port)
As a result on every server 2 ports should be opened to enable this traffic.
Load balancing
The Orleans runtime does all the load balancing for you. More important it is one of the pillars of the Orleans runtime. The runtime tries to make everything balanced, since balancing allows to maximize resource usage and avoid hotspots, which leads to better performance, as well as helps with elasticity.
More information: https://dotnet.github.io/orleans/Documentation/implementation/load_balancing.html
This means that it doesn’t make sense to put a load balancer between the clients and the cluster as this would defy one of the core purposes of the Orleans runtime.
But what if a load balancer exists?
This was exactly the situation I had with one of my customers. They were not fully aware that it doesn’t make sense to add a load balancer and installed our Orleans application on 2 servers behind a load balancer.
What they noticed was the following? As long as only one silo was running everything worked as expected. However the moment a second silo was spinned up, the other one shut down.
The problem was that due to the load balancer both silo got the same IP address. This was easily detectable in the Orleans Membershiptable:
To solve it, we explicitly specified the IP address of each server in the silo configuration: