This message was deleted.

Hi Abhinav, I know of a cluster running m5.large for Master and m5d.large for Query and Data ... it runs but doesn't have much capacity. So if you want to start really small and then grow node sizes as needed, this is one example. Thanks. John

Hi John,

As Druid rebuilds itself easily when nodes crash, you might try Spot instances to reduce costs if you can handle some moments with high latency. If you using AWS EKS, you can reduce costs removing Zookeeper and use a MiddleManager-less solution (no pre-allocated resources for MM workers, peons become k8s jobs). You mention you’re using i3.xlarge instances for data nodes, so your storage can go up to 900 gbs, dont need to keep in 50. You also don’t mention the size of your dataset, if you need live or batch ingestions, QPS… difficult to suggest anything like this.

What's your opinion on using a standalone installation for dev?

Again I think it is all dependent on the data size and workload. I have quickstart installations running on my Macbook with 8 worker task slots defined, but I use it only for demo and testing purposes. You can also run with K8s on a single instance. If your Dev environment is for functional testing only and not capacity/performance/stress testing, then this might be okay.