This message was deleted.

Hi Eddie, you can easily run a Druid cluster on one machine ... (I am running a test system on my Macbook pro) but the configuration should depend on what resources you have available. If you can list out the machines you have, -- i.e. how much RAM, cpu cores and disk space (and disk type) on each, if you are wiling to run k8s, and if your cluster needs to have High Availability -- that will make it easier to recommend a configuration.

ohhh really thanks i have same 5 machine cpu core 56, SCSI DISK, RAM 500G and don’t use kubernetest

Here's a starting suggestion: • Run Master Nodes on 3 machines (for HA) ... 8 cores and 16GB RAM each ◦ split between Overlord and Coordinator 50/50 ◦ you also have to run ZK on these • Run Query Nodes on 2 machines ... 8 cores and 16GB RAM each ◦ split between Broker/Router 75/25 • Run Data Nodes on all 5 machine ... 48 cores and 64GB each ◦ split between MM and Historical I'm not sure where to run RDS ... I don't normally run OSS Druid ... maybe someone else can chime in to adjust these splits.

Ah... I have already zk cluster with 4 machine used by original cluster and just changing zk.paths.base in druid config might be work! Thank you I would try and If I have other questions I would touch here again. Really thanks for follow up

Fyi there is a multi-level fan-out that happens in the architecture for query processing ... from the bottom up: • Data Node Processing Threads --> • Data Node Jetty Thread --> • Broker processing thread --> • Broker Jetty thread. 5 Historicals with 24 processing threads each might be a little bottom-heavy ... consider running 2 Historicals on each Data Node, so 10 Historicals with 12 processing threads each ... might be a bit more balanced of a hierarchy.

• Data Node(historical && mm) for all 5 machine ◦ 48 cores • overlord && coordinator for 3 machine • broker && router for 2 machine I have RDS mysql for metadata storage changing user name

druid.metadata.storage.connector.user

seems work thank you!! I would try and If I have other questions I would touch here again.

I believe you can run multiple of the same type of node on one machine, you may just need to use different port numbers. Read through the Tuning page: https://druid.apache.org/docs/latest/operations/basic-cluster-tuning.html ... there is a lot of important rule-of-thumb settings described here Historical

druid.processing.numThreads

... if you want to put two Historical on one node then give them 24 each. There is a lot of overassign of threads/processes to actual cpu cores so don't worry if your MM + Historical thread counts exceed the number of physical cores on the machine. That being said ... a question for you -- is your "56 cpu" count including hyperthreading? Because if not, then if you have hyperthreading that would bring your core count to 112 on each machine. Pay attention to http.numThreads, http.numConnections, and numMergeBuffers for capacity sizing on Broker and Historical. Heap sizes are also important ... don't make them too small or big. And different node types use different amounts of memory

Oh my God.. didn't know that changing port with two datanode in one machine. Like in one machine two historical node and two middlemanager. I thought in one machine one historical nodeband one middlemanager. Maybe because I'm using firewall I have been thought that for example overlord node port is 8090 something like that I would look into that maybe not hyperthread enabled but I would look into that too. I would setting druid config next week. Now setting for basic Linux setting. Thanks!!! Maybe when making cluster next week I think there will be bunches of questions. I hope I'm not in the way

Don't take my word for it ... I haven't tried it myself ... let me see if I can find confirmation on that ...

Oh I can try it because it's clean machine