How to limit dynamic self allocation of resources

Question

In our Hadoop cluster that runs under Yarn we are having a problem that some "smarter" people are able to eat significantly larger chunks of resources by configuring Spark jobs in pySpark Jupyter notebooks like:

conf = (SparkConf()
        .setAppName("name")
        .setMaster("yarn-client")
        .set("spark.executor.instances", "1000")
        .set("spark.executor.memory", "64g")
        )

sc = SparkContext(conf=conf)

This leads to the situation when these people literally squeeze out others less "smarter".

Is there a way to forbid users to self-allocate resources and leave resource allocation solely to Yarn?

Answer 1

YARN have very good support for capacity planning in Multi-tenancy cluster by queues, YARN ResourceManager uses CapacityScheduler by default .

Here we are taking queue name as alpha in spark submit for demo purpose.

$ ./bin/spark-submit --class path/to/class/file \
    --master yarn-cluster \
    --queue alpha \
    jar/location \
    args

Setup the queues:

CapacityScheduler has a predefined queue called root. All queues in the system are children of the root queue. In capacity-scheduler.xml, parameter yarn.scheduler.capacity.root.queues is used to define the child queues;

for example, to create 3 queues, specify the name of the queues in a comma separated list.

<property>
    <name>yarn.scheduler.capacity.root.queues</name>
    <value>alpha,beta,default</value>
    <description>The queues at the this level (root is the root queue).</description>
</property>

These are few important properties to consider for capacity planning.

<property>
    <name>yarn.scheduler.capacity.root.alpha.capacity</name>
    <value>50</value>
    <description>Queue capacity in percentage (%) as a float (e.g. 12.5). The sum of capacities for all queues, at each level, must be equal to 100. Applications in the queue may consume more resources than the queue’s capacity if there are free resources, providing elasticity.</description>
</property>

<property>
    <name>yarn.scheduler.capacity.root.alpha.maximum-capacity</name>
    <value>80</value>
    <description>Maximum queue capacity in percentage (%) as a float. This limits the elasticity for applications in the queue. Defaults to -1 which disables it.</description>
</property>

<property>
    <name>yarn.scheduler.capacity.root.alpha.minimum-capacity</name>
    <value>10</value>
    <description>Each queue enforces a limit on the percentage of resources allocated to a user at any given time, if there is demand for resources. The user limit can vary between a minimum and maximum value. The former (the minimum value) is set to this property value and the latter (the maximum value) depends on the number of users who have submitted applications. For e.g., suppose the value of this property is 25. If two users have submitted applications to a queue, no single user can use more than 50% of the queue resources. If a third user submits an application, no single user can use more than 33% of the queue resources. With 4 or more users, no user can use more than 25% of the queues resources. A value of 100 implies no user limits are imposed. The default is 100. Value is specified as a integer.</description>
</property>

links : YARN CapacityScheduler Queue Properties