36
Copyright ©2015 Treasure Data. All Rights Reserved. HDP2 and YARN operations point Ryu Kobayashi Treasure Data Tech Talk 11 and 12 Mar 2015

HDP2 and YARN operations point

Embed Size (px)

Citation preview

Page 1: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

HDP2 and

YARN operations pointRyu Kobayashi

Treasure Data Tech Talk 11 and 12 Mar 2015

Page 2: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Who am I?• Ryu Kobayashi • @ryu_kobayashi • https://github.com/ryukobayashi

• Treasure Data, Inc. • Software Engineer

• Background • Hadoop, Cassandra, Machine Learning, ... • I developed Huahin(Hadoop) Framework.

http://huahinframework.org/

Page 3: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

What is YARN?

Page 4: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

YARN(Yet Another Resource Negotiator) Architecture

Page 5: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

• MRv1

• JobTracker

• TaskTracker

Page 6: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

• YARN

• ResourceManager

• NodeManager

• ApplicationMaster

• Job History Server

• YARN Timeline Server

Page 7: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

• MRv1

• JobTracker

• TaskTracker

• YARN

• ResourceManager

• NodeManager

• ApplicationMaster

• Job History Server                                          (We  can  not  see  the  log  job  history  If  it  do  not  install)  

• YARN Timeline Server                                          (We  can  not  see  the  log  YARN  history  If  it  do  not  install)

Page 8: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

YARN Timeline Server • It is included container info

Page 9: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Note!!!

Page 10: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Use the Hadoop 2.4.0 and later!!!

Page 11: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

• The versions which must not be used

• Apache Hadoop 2.2.0

• Apache Hadoop 2.3.0

• HDP 2.0(2.2.0 based)

Page 12: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

• Currently

• Apache Hadoop 2.6.0

• CDH 5.3.2(2.5.0 based and patch)

• HDP 2.2(2.6.0 based and patch)

Page 13: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

• Why should not use?

• Capacity Scheduler

• There is a bug

• Fair Scheduler

• There is a bug

Page 14: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

• Any bugs?

• Each Scheduler will cause a deadlock

Page 15: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

•In fact, there is a bug in 2.4.0 and 2.6.0…

•It is better to use the new version.

•Note: 2.7.0 and later is a different thing

Page 16: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Backport Patch

• I was backport some patch

• https://github.com/ryukobayashi/patches

Page 17: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Backport Patch

• Included dead lock patch

• Format of the counter

• Application kill in Web UI.

Page 18: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Format of the counter

Page 19: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Format of the counter

Page 20: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Application kill in Web UI

Page 21: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Application kill in Web UI

Page 22: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Application kill in Web UI • Job kill in Web UI •

(default false)

• Application kill in Web UI •

(default true)

mapreduce.jobtracker.webinterface.trusted

yarn.resourcemanager.webapp.ui-actions.enabled

Page 23: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Backport Patch

• We want the next…

• Job task attempt kill in Web UI patch (in development)

• Currently, only command line

$ mapred job -kill-task attempt_*

Page 24: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources

• total container = 4

• concurrent application = 2

Page 25: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 2

Cluster  Application

App  Master Container

Application

App  Master Container

Page 26: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources

• total container = 4

• concurrent application = 4

Page 27: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 4

Cluster  Application

App  Master

Application

App  Master

Application

App  Master

Application

App  Master

Page 28: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 4

Cluster  Application

App  Master

Application

App  Master

Application

App  Master

Application

App  Master

Page 29: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 4

Cluster  Application

App  Master

Application

App  Master

Application

App  Master

Application

App  Master

Livelock!

Page 30: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 4

Cluster  Application

App  Master

Application

App  Master

Application

App  Master

Container

Application

App  Master

Kill

Page 31: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 4

• ^ squeeze the number of applications

Page 32: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 4

• ^ squeeze the number of applications

• set the root maxRunningApps

Page 33: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Matter of resources • total container = 4

• concurrent application = 4

• root maxRunningApps = 2

Cluster  Application

App  Master Container

Application

App  Master

Application

App  Master Container

Application

App  Master

Pending

Page 34: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

YANR Resource Managementyarn-site.xml

• yarn.nodemanager.resource.memory-mb • (yarn.nodenamager.vmem-pmem-ratio) • (yarn.scheduler.minimum-allocation-mb)

mapred-site.xml • yarn.app.mapreduce.am.resource.mb • mapreduce.map.memory.mb • mapreduce.reduce.memory.mb

fair-scheduler.xml • maxResources, minResources

etc…

Page 35: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

e.g. Use hdp-configuration-utils.py script http://goo.gl/L2hxyq

Use Ambari http://ambari.apache.org/

See the Cloudera’s document http://goo.gl/EBreca

YANR Resource Management

Page 36: HDP2 and YARN operations point

Copyright  ©2015  Treasure  Data.    All  Rights  Reserved.

Thanks!!!