I would like to do the documentation for this feature.

Gururaj Shetty Great. Feel free to ask us if you have any problem or questions.

Hi Tsuyoshi Ozawa ,
Thanks for replying ,can you please assign this issue to me . I would like to work on this.

I am planning to come up with a new page. Will provide the DRAFT soon. Let me know if you have any suggestion.

Just an information, all rmadmin commands are available in following page.
http://hadoop.apache.org/docs/r2.5.0/hadoop-yarn/hadoop-yarn-site/YarnCommands.html

Gururaj Shetty,
Thanks for volunteering to do this, but I've a WIP patch for this, would you mind me take over this task?

Wangda

And added this as sub task of YARN-2492.

Since the assignee is empty and got no response, taking over.

Wangda Tan Any update on documentation for Node Labels . Its difficult to completely evaluate this feature with out documentation

It would be good to add a documentation to ./hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/*.md.

Hi Tan, Wangda & Tsuyoshi Ozawa, i can help if anything is required for this ....

I've a doc for 2.6 and in apt format, will try to cover new changes in trunk and convert to markdown soon. Will keep you posted.

Attached initial version for review.

Fixed a bunch of formatting issues and reattached patch.

+1 overall

This message was automatically generated.

Hi Wangda Tan
Had a few points from my side (many are cosmetic than technical, you can check whether they are required as its my view)

1. Regarding Exclusive and Non-exclusive, could it be more explicit/better word it as, Exclusive Partition doesnt share the resources with containers which is not requesting for the same partition(label). and Non Exclusive shares the resources with containers which are requesting for Default partition also .
2. As per YARN-2498, when we need to preempt containers from a queue's partition, we will first preempt ignore_partition_exclusivity allocated containers.
   this needs to be captured in the doc, And (not related to this jira)also would it be good to provide an option to not allocate containers in non exclusive partition as the containers can get preempted ?
3. Concept of Default Partition needs to be explained (if no labels are assigned then it belongs to default partition or something ... and requests which are not specifying any label will be given for Default Partition). "Default_Partition" is used in the Web page hence better captured in the doc too
4. one application can only use subset of node labels can be accessed by the queue which contains the application => one application can only use subset of node labels that can be accessed by the queue which contains the application
5. You can specify which node partition is need when allocating container or submiting application => You can specify which node partition is needed when allocating container or submiting application. or just You can specify required node partition when allocating container or submiting application.
6. Percentage of each queue can access resource with each label => Specify percentage of resource of a partition which can be accessed by a queue
7. You can specify can it be user can specify, i feel using of "I" and "you" in documentation would not be good
8. Specify node label requirement in Resource request, => Specify required Node Label in Resource request
9. Update node labels - user/admin can update labels => i think only admin can update using yarn rmadmin command
10. value mentioned for yarn.node-labels.fs-store.root-dir & yarn.node-labels.enabled looks like the default value which i think is not ? so may be as in other Property tables better to describe about the property and the default value. Also would it be better to specfiy that local file system can also be configured ?
11. yarn rmadmin -addToClusterNodeLabels label_1(exclusive=true/false),label_2(exclusive=true/false) => yarn rmadmin -addToClusterNodeLabels " label_1(exclusive=true/false),label_2(exclusive=true/false) "
12. When there’re labels on some nodes in the cluster, this means percentage of resources on nodes without label. dint get the meaning of this sentence
13. If you want to explicitly specify a queue can only access nodes without labels, just put a space as the value. even blank or no configuration has the same effect right ?
14. Set the percentage of the queue can access to nodes with <label>. => Set the percentage of resource of a label which can be accessed by a queue
15. maximum-capacity default value is 100 right ? & default for capacity is 0 ?
16. Value like “hbase”, which means: if applications submitted to the queue without specifying node label in their resource requests, it will use the default-node-label-expression => better place the end part as it will use "hbase" as default-node-label-expression.
17. (I used key=value here for readability) => (key=value is used here for readability)) and in the example total capacity of all children queues @ root is 99
18. No mention about the Node Label UI or about the REST API for Node Labels ? RESt would be better captured in another page . thoughts ?
19. Would it be better to give a link in CS web doc page mentioning that Node Labels scheduler Configuration can be referred to this node labels doc's section ?

Hi Naganarasimha G R,
Thanks for your thoughtful review, comments for your suggestions:
2) There's no preemption related documentation in Apache Hadoop yet, I suggest to add this part after we have a preemption page.
10) They're what admin should specify. I prefer to not add default value here because default is always changing, which will be tracked by yarn-default.xml
12) Changed it, it should be percentage of resources on nodes with DEFAULT partition.
13) That's different, <value></value> and not specifed means "inherit from parent"
18) REST API is under development, I think we still need some time to finalize it for 2.8. I suggest to add that part later.
19) Added CS link from node labels page, I think it's a relative independent feature. I suggest to not reference from CS.

I addressed other items in attached patch.

Please let me know your ideas.

Thanks,

-1 overall

This message was automatically generated.

hi Wangda Tan, seems like after applying the patch mvn site is failing

Hi Wangda Tan,
After escaping the links, seems like its getting applied. Few nits :

• User need configure how many resources => User need configure how much resource of each partition
• points in note after configuration section needs to come as list
• application can use following Java APIs => Application can use following Java APIs

Apart from it others seems to be fine !

Thanks Naganarasimha G R for additional review, attached ver.3 patch.

+1 overall

This message was automatically generated.

Hi Wangda Tan, can we get this patch committed ? anything else pending for this ?

I'll check this soon.

Naganarasimha G R Wangda Tan Could you fix following points?

1. Could you edit hadoop-project/src/site/site.xml to add an entry of NodeLabel to the left-side menu?
2. there are some typos:

+Yarn Node Labels

"Yarn" should be upper case, YARN?

and application can specify where to run.

applications instead of application looks better.

User need configure how much resources of each partition can be used by different queues, see next sections.

"User need to configure" looks better. Also, "~ by different queues, see next sections" is a bit difficult to read. My suggestion is "~ by different queues. For more detail, please see the next section."

There’re two kinds of node partitions

"’" should be "'"(using half-width character instead of full-width character). Adding : looks better after the sentence.

Exclusive: Containers will be allocated to nodes with exactly match node partition.

Containers should be containers.

User can specify ~

A user can specify ~

user can set percentage like: queue-A can access 30% ~

a user can set ~, adding space between queue and , and A.

So each of they can use 1/3 ~

each of them ~

After finish setting configuration of CapacityScheduler, executing ~

After finishing configuration of CapacityScheduler, execute ~

Application can use ~

Applications can use ~

Thanks Tsuyoshi Ozawa, for the comments. Wangda Tan as it was pending for a while and it was only small corrections, also i had to update document for distributed NodeLabels on top of this hence i have given a patch for this jira.
Tsuyoshi Ozawa, have corrected most of your comments but just for few where in user was addressed in singular " A User", as i felt existing was fine.

+1 overall

This message was automatically generated.

Sorry I missed this comment, thanks Naganarasimha G R addressing them and review from Tsuyoshi Ozawa!

+1, committing this shortly.

FAILURE: Integrated in Hadoop-trunk-Commit #8378 (See https://builds.apache.org/job/Hadoop-trunk-Commit/8378/)
YARN-2801. Add documentation for node labels feature. Contributed by Wangda Tan and Naganarasimha G R. (ozawa: rev faa38e1aa49907254bf981662a8aeb5dc52e75e3)

• hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/NodeLabel.md
• hadoop-yarn-project/CHANGES.txt
• hadoop-project/src/site/site.xml

FAILURE: Integrated in Hadoop-Yarn-trunk-Java8 #334 (See https://builds.apache.org/job/Hadoop-Yarn-trunk-Java8/334/)
YARN-2801. Add documentation for node labels feature. Contributed by Wangda Tan and Naganarasimha G R. (ozawa: rev faa38e1aa49907254bf981662a8aeb5dc52e75e3)

• hadoop-yarn-project/CHANGES.txt
• hadoop-project/src/site/site.xml
• hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/NodeLabel.md

Committed this to trunk, branch-2, and branch-2.7. Thanks for your contributions, Wangda Tan and Naganarasimha G R!

FAILURE: Integrated in Hadoop-Yarn-trunk #1061 (See https://builds.apache.org/job/Hadoop-Yarn-trunk/1061/)
YARN-2801. Add documentation for node labels feature. Contributed by Wangda Tan and Naganarasimha G R. (ozawa: rev faa38e1aa49907254bf981662a8aeb5dc52e75e3)

• hadoop-project/src/site/site.xml
• hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/NodeLabel.md
• hadoop-yarn-project/CHANGES.txt

SUCCESS: Integrated in Hadoop-Mapreduce-trunk-Java8 #326 (See https://builds.apache.org/job/Hadoop-Mapreduce-trunk-Java8/326/)
YARN-2801. Add documentation for node labels feature. Contributed by Wangda Tan and Naganarasimha G R. (ozawa: rev faa38e1aa49907254bf981662a8aeb5dc52e75e3)

• hadoop-yarn-project/CHANGES.txt
• hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/NodeLabel.md
• hadoop-project/src/site/site.xml

FAILURE: Integrated in Hadoop-Mapreduce-trunk #2275 (See https://builds.apache.org/job/Hadoop-Mapreduce-trunk/2275/)
YARN-2801. Add documentation for node labels feature. Contributed by Wangda Tan and Naganarasimha G R. (ozawa: rev faa38e1aa49907254bf981662a8aeb5dc52e75e3)

• hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/NodeLabel.md
• hadoop-yarn-project/CHANGES.txt
• hadoop-project/src/site/site.xml

FAILURE: Integrated in Hadoop-Hdfs-trunk #2257 (See https://builds.apache.org/job/Hadoop-Hdfs-trunk/2257/)
YARN-2801. Add documentation for node labels feature. Contributed by Wangda Tan and Naganarasimha G R. (ozawa: rev faa38e1aa49907254bf981662a8aeb5dc52e75e3)

• hadoop-project/src/site/site.xml
• hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/NodeLabel.md
• hadoop-yarn-project/CHANGES.txt

FAILURE: Integrated in Hadoop-Hdfs-trunk-Java8 #318 (See https://builds.apache.org/job/Hadoop-Hdfs-trunk-Java8/318/)
YARN-2801. Add documentation for node labels feature. Contributed by Wangda Tan and Naganarasimha G R. (ozawa: rev faa38e1aa49907254bf981662a8aeb5dc52e75e3)

• hadoop-yarn-project/CHANGES.txt
• hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/NodeLabel.md
• hadoop-project/src/site/site.xml

Thanks Tsuyoshi Ozawa and Naganarasimha G R!