hadoop-hdfs-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Naganarasimha G R (Naga)" <garlanaganarasi...@huawei.com>
Subject RE: Question about YARN Memory allocation
Date Wed, 28 Jan 2015 08:35:21 GMT
Hi Jungtaek Lim,
Earlier we faced similar problem of reservation with Capacity scheduler and its actually solved
with YARN-1769 (part of 2.6 hadoop)
So hope it might help you if you have configured Capacity scheduler. also check whether "yarn.scheduler.capacity.node-locality-delay"
is configured (might not be direct help but might reduce probability of reservation ).
I have one doubt with info : In the image it seems to be 20GB and 10 vcores reserved but you
seem to say all are reserved ?
Is LCE & DRC also configured ? if so what are the vcores configured for NM and the app's
containers?

Regards,
Naga
________________________________
From: 임정택 [kabhwan@gmail.com]
Sent: Wednesday, January 28, 2015 13:23
To: user@hadoop.apache.org
Subject: Re: Question about YARN Memory allocation

Forgot to add one thing, all memory (120G) is reserved now.

Apps Submitted  Apps Pending    Apps Running    Apps Completed  Containers Running      Memory
Used     Memory Total    Memory Reserved VCores Used     VCores Total    VCores Reserved Active
Nodes    Decommissioned Nodes    Lost Nodes      Unhealthy Nodes Rebooted Nodes
2       1       1       0       60      120 GB  120 GB  20 GB   60      80      10      10
     0       0       0       0

Furthermore, 10 more VCores are reserved. I don't know what is it.


2015-01-28 16:47 GMT+09:00 임정택 <kabhwan@gmail.com<mailto:kabhwan@gmail.com>>:
Hello all!

I'm new to YARN, so it could be beginner question.
(I've been used MRv1 and changed just now.)

I'm using HBase with 3 masters and 10 slaves - CDH 5.2 (Hadoop 2.5.0).
In order to migrate MRv1 to YARN, I read several docs, and change configrations.

```
yarn.nodemanager.resource.memory-mb: 12288
yarn.scheduler.minimum-allocation-mb: 512
mapreduce.map.memory.mb: 1536
mapreduce.reduce.memory.mb: 1536
mapreduce.map.java.opts: -Xmx1024m -Dfile.encoding=UTF-8 -Dfile.client.encoding=UTF-8 -Dclient.encoding.override=UTF-8
mapreduce.reduce.java.opts: -Xmx1024m -Dfile.encoding=UTF-8 -Dfile.client.encoding=UTF-8 -Dclient.encoding.override=UTF-8
```

I'm expecting that it will be 80 containers running concurrently, but in real it's 60 containers.
(59 maps ran concurrently, maybe 1 is ApplicationManager.)

All YarnChilds' VIRT are higher than 1.5G and lower than 2G now, so I'm suspecting it.
But it's better to make clear, to understand YARN clearer.

Any helps & explanations are really appreciated.
Thanks!

Best regards.
Jungtaek Lim (HeartSaVioR)




--
Name : 임 정택
Blog : http://www.heartsavior.net / http://dev.heartsavior.net
Twitter : http://twitter.com/heartsavior
LinkedIn : http://www.linkedin.com/in/heartsavior
Mime
View raw message