This message was deleted.
# general
s
This message was deleted.
p
2 of existing VMs for brokers 2 of existing VMs for overlord+coordinator + router (for HA) all of additional VMs for historicals. You didnt mention current middlemanagers
g
Hello now we have 2 datanode that have middlemanager and historical 1 master that have coordinator-overlord 1 query that have broker and router
if in new cluster we have: 1 master node with coordinator-overlord 2 query nodes with brokers and routers 6 datanodes with middlemanager and historical?
p
dont mix historicals with middle
Each query needs to process X segments from historicals(and peons if realtime), one segment can be processed only by one cpu.
g
hmm ok lets see what will be the best scenario for 10 VMs, each one with 18GB memory, 4 cores and 300GB storage (All changes accepted). 2 VMs for brokers 2 VMs for coordinator-overlord and routers 2 VMs for middlemanager 4 VMs for historical ??
p
coordinator-overlord and routers are not resource-hungry components, personally i would try sqeeze them maybe with brokers, what would give me 2 more historicals
is 300GB storage is dedicated or shared?
g
dedicated! we use a platform for visualization that query a lot the druid data every day. so we want cache and we want response time to be as fast as can.
also we will have 1 more VM for deep storage (HDFS) as we have now in as-is
In conclusion, I have decided to proceed as follows and do a first test: 2 VMs for query node that will have brokers and routers 32GB RAM , 8 cores , 50GB storage 1 VM for master node coordinator-overlord 18GB RAM , 4 cores, 50GB storage 2 VMs for datanode only with middlemanager 18GB RAM , 4 cores, 300 GB storage 6 VMs for datanode only with historical 32GB RAM , 8 cores, 300 GB storage What's your opinion?
p
sounds good.