Hey, We want to go production with pinot what is t...
# troubleshooting
r
Hey, We want to go production with pinot what is the recommended approach for deploying pinot... 1. Deploying all the components in one single instance 2. Deploying each component Controller, Broker and Server on different instances with different cores and memory provided(Based on load). 3. Deploying each component separately on EKS which can be costly but we can deploy it safely 4. Preferred instance type (at component level if standalone or at Kubernetes node level) a. M / R or C series instance types (We are on AWS cloud) Any suggestion? @Xiang Fu @Mayank C.C: @Kavya
x
To make the system reliable, we suggest to deploy with separated components, e.g. 3 zookeeper, 2+ controller, 2+ brokers, 2+ servers, etc for EKS, depends on your node-selector and resources(request/limit), your initial cost will be higher, but later on you just need to horizontally scale servers when you have more data
k
Hi @Xiang Fu, which one would be better ? to deploy these components in separate EC2 Machines OR to have one EKS and deploy all components in the EKS?
x
we usually recommend to use M/T series for zk/controller/broker, For servers, we usually prefer larger memory ones like R-series. However if your team ever has the ability to manage NVME localSSD, we would recommend to use i series, which is also what we internal offer as managed solution for our users.
EKS is better manageable than EC2
It really depends on your teams technical and operational depth on Pinot and K8s
k
Okay, thanks for the details. Also what is the recommended minimum cpu and memory for each of the components - ZK, Controller, Broker and Server? Any concern if we deploy all components in single EKS machine?
x
it depends on your cluster size I would say, we usually have 2 cpu for zk/controller/broker at starting point and scale based on their usage
note, if you deploy on a single ec2 instance for a EKS cluster, then you have to sacrifice the HA in certain disaster scenarios