How to optimally set spark driver properties in YARN - amazon-web-services

Question

I am trying out various options for setting spark driver memory in yarn.
Use Case:
I have a spark cluster with 1 master and 2 slaves
master : r5d xlarge - 8 vcore, 32GB
slave : r5d xlarge - 8 vcore, 32GB
I am using Apache Zeppelin to run the queries on spark cluster. Spark interpreter is configured with properties provided by Zeppelin. I am using spark 2.3.1 running on YARN. I want to create 4 interpreters so that 4 users can parallelly use this cluster.
Config 1:
spark.submit.deployMode client
spark.driver.cores 7
spark.driver.memory 24G
spark.driver.memoryOverhead 3072M
spark.executor.cores 1
spark.executor.memory 3G
spark.executor.memoryOverhead 512M
spark.yarn.am.cores 1
spark.yarn.am.memory 3G
spark.yarn.am.memoryOverhead 512M
Below is the spark executor UI:
Config 2:
spark.submit.deployMode client
spark.driver.cores 7
spark.driver.memory 12G
spark.driver.memoryOverhead 3072M
spark.executor.cores 1
spark.executor.memory 3G
spark.executor.memoryOverhead 512M
spark.yarn.am.cores 1
spark.yarn.am.memory 3G
spark.yarn.am.memoryOverhead 512M
Below is the spark executor UI:
Questions:
Why is the container size of driver 0?
Is the spark.memory.fraction calculated as (spark.driver.memory-300)*0.6 ? If so, why is it not exact ? (14.22, 7.02 resp)
Why is the container size of executor 3.8 GB ? According to my configuration, it should be 3G + 512M = 3.5 GB. This issue was not there with spark 2.1
No of VCores available to YARN is 8 per node. How is this possible since AWS gives vCPU with their instances? Hence I should only be getting 4 VCores according to AWS.
https://aws.amazon.com/ec2/instance-types/r5/
If I want to use 4 interpreters, should I distribute 32 GB of master equally to all the interpreters?
Driver:
spark.driver.cores 2
spark.driver.memory 7G
spark.driver.memoryOverhead 1024M