Look at the following figure,Cluster Utilization is low, but Yarn is high.
Anyone help? thanks.
Have you changed settings for yarn resources?
You can set YARN resources by editing the yarn-site.xml file related to RAM and processors:
The first two are for the ResourceManager, the last two refer to each node. Remember to leave some RAM and CPU for other services on your machines!
Taken from Yarn showing low number of cores in Mapr Control Center
Thanks for your help.
I really have not changed settings for yarn resources.
I will update the question after change the setting.
Hi Mathieu Dumoulin,
I have changed settings for yarn resources like this:
then Yarn console show this:
So i hava some question:
1.why Total Memory is 497? I set yarn.nodemanager.resource.memory-mb=204800, it is 200G, Why Total Memory is not 600G? Also CPU'S Total have the same question.
2.Used Disks is 0 ? I am sure i had use the disks .
Before saying anything, lets get the basics out of the way:
1. did you copy the yarn-site.xml file to all nodes of the cluster?
2. did you restart warden (at least the resource manager nodes, but all nodes to be safe if it's not a production cluster)
Your other questions:
1. Keep in mind that YARN is sharing resources with other system services. You may need to look at warden configuration to see how much total resources are allocated for YARN in the first place, it's possible you're hitting the maximum that warden is giving you. It's not guaranteed you will get all the resources you ask for. Test if this is the case by trying settings that are very small, like 4 vcores and 10000 MB memory per NM, and you should see the exact numbers after you restart warden.
2. Number of Disks used for MapR, zero (0) is normal. MapR-FS doesn't work like YARN on HDFS and so these settings are specific to MapR. you wouldn't need to normally worry about this.
A great resources to help you with YARN resource configuration on MapR: Best Practices for YARN Resource Management | MapR
For the sake of not talking without checking first, I have tested this on a freshly installed 5.2.1 cluster running on AWS EC2 instances.
With default settings, this is what I see in MCS:
I changed the settings like so:
And copied the file yarn-site.xml to all nodes and restarted warden. MCS shows the following:
Exactly the settings I used.
Please double check you are indeed following these steps properly.
That's great.I restart the cluster environment. Now the yarn console display is normal.
I set yarn.scheduler.maximum-allocation-mb=16G,and now just one application is running,why it use 141G menory? Should not be 16 G?
Hi, Zou.No, it shouldn't.
The property yarn.scheduler.maximum-allocation-mb is responsible for the maximum amount of RAM that can be allocated to a single container in YARN (mapper or reducer).So, for example, if you have 3 Node Managers and each has 2 containers running - you'll have 6 containers in total, which will show 96 Gb of RAM utilized by this application (3 NM * 2 containers * 16 RAM for each, plus some memory is also used by Application Master container). In your case number 141 means that application of yours is using 141 Gb of RAM in sum with all its containers.
You can easily check the cluster metrics and container allocation for specific application in Resource Manager User Interface:
http://<RM_hostname>:8088 (for unsecured cluster)
https://<RM_hostname>:8090 (for secured cluster)
And then you would be able to navigate through applications and their containers to check any detailed info.
Thank you, through your explanation, I understand.
Retrieving data ...