We are using ElastiCache for Redis with cluster mode disabled, scaled up for all 5 nodes using cache.m6g.large, connecting to the reader endpoint of the cluster from a multi-instance Node.js application running in ECS. Application instances seem to be assigned to reader nodes randomly, resulting a random distribution of load between the nodes. This works out as an even distribution sometimes, some other times one or two nodes cap out at 100% engine CPU utilization (crashing our application) while others run idle. Please advise, how should we distribute connections evenly at all times?
AWS 官方已更新 2 年前