Support Questions

Find answers, ask questions, and share your expertise
Announcements
Celebrating as our community reaches 100,000 members! Thank you!

Swap Memory Usage for HDFS, Yarn and Impala

avatar
Expert Contributor

Hello,

 

We have been getting regular warning messages of Swap Memory being utilized more than the threshold limit.

 

Currently, we have set swap memory threshold as below:

For,

HDFS = 100 MB

Impala = 30 MB

Yarn = 500 MB (Could be +-20%, I dont remember the exact number).

 

Swap Memory usage for each of the component is crossing the threshold and reaches upto 720 MB (in case of Yarn). Because of this, we usually see warnings on our CM dashboard.

 

I know increasing the Swap Memory Threshold could remove these warnings, however, we would prefer rather reducing the usage.

 

Would be great if anyone could suggestion any Memory Tuning options which would reduce the usage of Swap Memory. Also, if there is a best recommendations of setting memory usage threshold, kindly share that as well.

 

Thanks

Snm

7 REPLIES 7

avatar
Explorer

Hi,

 

did you set the swappiness in the OS?

 

Best

Andy

avatar
Expert Contributor
Hello Vogone,

Thanks for the reply.

Yes, swappiness is set to 1.

Regards,
snm

avatar
Master Guru

@snm1523,

 

Cloudera recommends the following:

 

https://www.cloudera.com/documentation/enterprise/latest/topics/cdh_admin_performance.html#cdh_perfo...

 

Essentially, we don't want swapping to occur.

avatar
Expert Contributor
Hello @bgoogley,

Thank you for the reply.

We have already set the swappiness to 1. Still getting these swappiness warnings.

Regards,
Snm

avatar
New Contributor

@snm1523,

 

Disk IO intensive workloads can cause the kernel to swap even if swappiness is set to 1.

 

Have you tried increasing the vm.vfs_cache_pressure kernel parameter on all worker nodes? Here is a post on how to fine tune the parameter: http://datavelo.com/en/2018/04/10/kernel-swapping-vm-swappiness-1

 

Best,

z

avatar
Rising Star

Any working solution for the swap problem?

 

We have same problem on CentOS 7 with CDH Enterprise 5.14.0

 

DataNode, NodeManager, YarnChild processes are swapping

avatar
Explorer

As a temporary fix you can use linux command swapoff -a && swapon -a to move swap. Just make sure you have enough free memory to move swap to (top command)