[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-14 Thread Rong Rong (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16907465#comment-16907465
 ] 

Rong Rong commented on FLINK-13548:
---

Thanks for the feedback [~till.rohrmann]. Yes I think so too. And since this is 
only a configuration key-value pair. it will simply just get ignored for older 
version. I've run through flink-yarn-test against both new and old version of 
Hadoop YARN and they all look pretty promising. I will run some more tests 
before merging it. 

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: boxiu
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is based on YARN official doc.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> priority value is non-negative, the bigger the number, the higher the 
> priority.  By default, we take -1. When the priority is negative, we use 
> default yarn queue priority.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-14 Thread Till Rohrmann (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16907116#comment-16907116
 ] 

Till Rohrmann commented on FLINK-13548:
---

I think we could add it to Flink. We should simply document that this feature 
only works with Hadoop >= 2.8. If there are API calls which are not available 
in earlier versions, then we could solve similar to 
https://github.com/apache/flink/blob/master/flink-yarn/src/main/java/org/apache/flink/yarn/RegisterApplicationMasterResponseReflector.java.

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: boxiu
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is based on YARN official doc.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> priority value is non-negative, the bigger the number, the higher the 
> priority.  By default, we take -1. When the priority is negative, we use 
> default yarn queue priority.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-11 Thread Rong Rong (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16904697#comment-16904697
 ] 

Rong Rong commented on FLINK-13548:
---

Hi [~boswell] thanks for the contribution. 
I just come to realize that the priority scheduling feature was only there in 
[YARN 2.8.x and 
up|https://hadoop.apache.org/docs/r2.8.5/hadoop-yarn/hadoop-yarn-site/CapacityScheduler.html#Setup_for_application_priority.].
 
Although we can build Flink with Hadoop 2.8.x + using [specific 
commands|https://ci.apache.org/projects/flink/flink-docs-stable/flinkDev/building.html#hadoop-versions].
 I was wondering if we should implement a feature that only works in one 
version. 

CC [~till.rohrmann] who might have better insight here. 

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: boxiu
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is based on YARN official doc.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> priority value is non-negative, the bigger the number, the higher the 
> priority.  By default, we take -1. When the priority is negative, we use 
> default yarn queue priority.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-07 Thread thy (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16901753#comment-16901753
 ] 

thy commented on FLINK-13548:
-

ok, I currently have pushed a new commit according to the last discussion~

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: thy
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is based on YARN official doc.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> priority value is non-negative, the bigger the number, the higher the 
> priority.  By default, we take -1. When the priority is negative, we use 
> default yarn queue priority.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-06 Thread Rong Rong (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16901356#comment-16901356
 ] 

Rong Rong commented on FLINK-13548:
---

I think when you submit the request for priority that is out of range YARN 
should return some error/warnings but I am not sure. I think it is fine for now 
to introduce this first and refine later with the sanity check.

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: thy
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is 1.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> submission of the priority range from 1 to 9(from low to high priority). When 
> the user does not configure the parameters, the priority is still 1. 



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-06 Thread thy (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16900977#comment-16900977
 ] 

thy commented on FLINK-13548:
-

[~walterddr]: Yes, it is better to do a sanity check. Do you have any good idea 
to get the yarn priority info?

Thanks~

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: thy
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is 1.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> submission of the priority range from 1 to 9(from low to high priority). When 
> the user does not configure the parameters, the priority is still 1. 



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-05 Thread Rong Rong (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16900232#comment-16900232
 ] 

Rong Rong commented on FLINK-13548:
---

Yes. I think setting it to the default YARN priority (of the queue submitting 
to) is probably the right way to do. Implementation details on whether to use a 
flink conf mapping can be discuss in the PR; 
Regarding the values for the priority, I think setting it to a non-negative 
number is fine for now. but I was wondering if we need to sanity check the 
values against the cluster/queue setting? IMO it is problematic submitting 
Flink application using an out-of-range priority number - even YARN interprets 
them correctly to the max value, we should at least throw a warning (as this 
behavior might not be intended by the user)



> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: thy
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is 1.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> submission of the priority range from 1 to 9(from low to high priority). When 
> the user does not configure the parameters, the priority is still 1. 



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-05 Thread thy (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16900099#comment-16900099
 ] 

thy commented on FLINK-13548:
-

[~walterddr]: Thanks for your advice!
First, I would say sorry that I made a mistake between our internal spark and 
community spark. The internal spark is implemented as what I talked above. 
Actually the community spark does not implement the yarn application priority. 
1. As mentioned in YARN official doc about "Cluster-level priority": Any 
application submitted with a priority greater than the cluster-max priority 
will have its priority reset to the cluster-max priority. I think maybe it is 
not necessary to limit 1-9, just make sure that the priority is not negative.
2. The queue’s default priority will be used for any application submitted 
without a specified priority. An idea to envision the default value is that we 
set the default YARN PRIORITY OPTION to -1. Once we get -1 from the option, we 
do not set "appcontext.setPriority". Maybe there is a better choice?

Thanks

thy

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: thy
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is 1.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> submission of the priority range from 1 to 9(from low to high priority). When 
> the user does not configure the parameters, the priority is still 1. 



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)


[jira] [Commented] (FLINK-13548) Support priority of the Flink YARN application

2019-08-04 Thread Rong Rong (JIRA)


[ 
https://issues.apache.org/jira/browse/FLINK-13548?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16899652#comment-16899652
 ] 

Rong Rong commented on FLINK-13548:
---

Hi [~boswell], thanks for bringing up this discussion. I think this is indeed a 
good feature to have. Several questions regarding the idea here: 
1. do we need to limit the priority to 1-9? According to  [YARN official 
doc|https://hadoop.apache.org/docs/current/hadoop-yarn/hadoop-yarn-site/CapacityScheduler.html#Setup_for_application_priority],
 seems like we can actually customize how these priority integers are 
interpret. 
2. how do you envision the default value should be interpret? I think this 
should probably be a default YARN setting instead of a hard-coded value in 
flink-conf.yaml. But I am not sure whether there's a right way to do it. maybe 
you can share how Spark does it based on your experience. 
Thanks,
Rong

 

> Support priority of the Flink YARN application
> --
>
> Key: FLINK-13548
> URL: https://issues.apache.org/jira/browse/FLINK-13548
> Project: Flink
>  Issue Type: Improvement
>  Components: Deployment / YARN
>Reporter: thy
>Priority: Major
>  Labels: pull-request-available
>  Time Spent: 10m
>  Remaining Estimate: 0h
>
> Currently, Flink 1.9 does not support yarn priority submission. The default 
> priority of yarn submission jobs is 1.
> Based on this, we can provide a ConfigOption in YarnConfigOptions. The 
> submission of the priority range from 1 to 9(from low to high priority). When 
> the user does not configure the parameters, the priority is still 1. 



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)