I'm testing out a small Redis cache on ECS Fargate and I've hit a wall with task-level sizing limitations. In Fargate, tasks require CPU and memory to be set according to specific approved combinations. Although I can adjust CPU and memory at the container level, the total still must conform to these combos. My Redis workload primarily involves reading with light writing, peaking at about 0.5 vCPU, but it requires around 5 GB of RAM due to how Redis manages memory. Unfortunately, Fargate only permits a maximum of 4 GB of RAM for 0.5 vCPU, so to get the 5 GB I need, I'd have to opt for 1 vCPU, which feels like paying for double the CPU I won't use. I can't lower the RAM without risking memory issues with Redis. It seems like AWS's strategy here is aimed at scheduling efficiency, but this setup feels like forced over-provisioning for CPU-light, memory-heavy workloads. Has anyone experienced this? Should I just accept the increased costs, switch to ECS on EC2 for flexibility, or is there another Fargate strategy I might be overlooking?
2 Answers
If you're looking for better control over your memory-to-vCPU ratios, moving to ECS on EC2 might be your best bet. Fargate's limitations can be pretty restrictive for workloads like yours.
Could this be something to bring up as a feature request? It seems a lot of users could benefit from more flexible sizing.
It might not be an intentional hassle from AWS; they likely find it more cost-effective to maintain fixed size limits for scheduling. In the grand scheme, the difference in pricing could be minimal if you run the container continuously. Just weigh the cost of extra cloud expenses against the time you'd spend trying to fix this—might just be worth it in the end!
Definitely, for a single container it isn't massive, but for multiple workloads it can add up quickly. Just wish Fargate had better handling for these CPU-light, memory-heavy scenarios!
I heard some services have specific limits too; for example, Lambda has short-duration tasks. Seems like EC2 might be the way to go if Fargate's limits are a dealbreaker.