Servers start with a 50GiB EBS Volume

[#1]

All servers seem to be starting with a 50GiB EBS volume as the root device. Is it possible to change this to just use ephemeral storage on nodes that have substantial ephemeral storage. Below is a picture i took from my 4 node cluster

[#2 · Accepted solution]

Hi Chris,

If you would like to launch an instance on AWS you have basically two options for root volumes:

• Instance Store backed root volumes
• EBS backed root volumes

So you cannot really use ephemeral storage as root disks. In conceptional point of view the Instance Storage is very similar to ephemeral one, but Cloudbreak does not support it mainly for the following reasons:

1. Instance Store has slower boot time it is up to 5 mins, the esb is up to 1 mins according to specification.
2. Instance Store has a limit in size which is 10GB which does not really make suitable for installing HDP packages on it
3. a lot of base ami support EBS root disk only, e.g. I was not able to find Redhat 7.2 base ami with “Instance Store” support

You can read more about root disk storage on AWS site:

https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ComponentsAMIs.html#storage-for-the-root-device

Cloudbreak supports EBS only and with a fixed size of 50GB, although this 50GB is not configurable at the moment, but based on a feature request we can make it configurable.

The ephemeral disks selected by you on Cloudbreak UI is used for HDFS data, so Cloudbreak puts the HDFS data on those ephemeral disks, but all HDP RPMs will be installed on the root disk with is EBS backed.

Attila

[#3]

Thanks a ton. This really clarifies things for me.

One side question, just for my own understanding, the fixed 50gb volumes are for non hdfs storage only right? If i understand you hdfs does not go on there.

[#4]

Yes, the 50gb fixed root disk is for OS and RPMs, configs etc., but not for HDFS.