We created a simple load balancer on AWS with the target group being 3 ec2 instances.
The game flow is usually: send rpc request to create match, then join the match and wait for opponent to join.
However when the load balancer is active, even though the rpc requests succeed for creating a match, the join_match fails with the returned match ID, indicating that the matches are not being synced across instances when the load balancer requests are being sent properly.
What is the suggested solution for this scenario? Should we save the matches object to the database server?
@GroovyAntoidDev For the scenario where you have multiple Nakama instances you want clustered together and have it auto-balance matches across servers you should use the Managed Cloud service. It helps our business and enables us to continue to make as much of our technology open-source as possible.
Thank you I guess this answers my question 
We will definitely contact you for the managed cloud service in the future.
As a side question, is it possible to have multiple nakama instances clustered without auto-balancing matches across them?
I noticed my earlier comment is vague and can be misunderstood, what I meant was: is it possible to set up the multiple instances to respond to rpc requests properly for each user (such as for coin reward requests) even if multiplayer matches are not balanced?