You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Bibin Chundatt (Jira)" <ji...@apache.org> on 2020/07/02 04:46:00 UTC

[jira] [Comment Edited] (YARN-10335) Improve scheduling of containers based on node health

    [ https://issues.apache.org/jira/browse/YARN-10335?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17149843#comment-17149843 ] 

Bibin Chundatt edited comment on YARN-10335 at 7/2/20, 4:45 AM:
----------------------------------------------------------------

Thank you for showing interest in the JIRA [~cyrusjackson25]

Adding what i have in mind about the health detail. Node manager  has node health service which returns a boolean value .Sends UNHEALTHY if the node health script return error / If  we don't have any healthy local  directories. 

We will introduce field/fields which returns detailed node health value about the node along with the NodeHealthStatus.  

Example:
{quote}
message NodeHealthStatusProto {
    optional bool isHealthy = 1;
    optional string nodeHealthDescription = 2;
    optional string exceptionString = 3;
    optional NodeHealthDetail nodehealthDetail=4;
    optional StringIntMapProto nodeHealthdetail=5;
}

message StringStringMapProto {
  optional string key = 1;
  optional int32 value = 2;
}

keys could be - overall , ssd, non ssd, etc.. 
{quote}

Also make the NodeHealthService pluggable to support custom implementations of NodeHealthServices.


was (Author: bibinchundatt):
Thank you for showing interest in the JIRA [~cyrusjackson25]

Adding the thought what i have in mind about the health value. Node manager  has node health service which returns a boolean value . 
Sends UNHEALTHY if the node health script return error / If  we don't have any healthy local  directories. 

We want to introduce field/fields which returns detailed node health value about the node along with the NodeHealthStatus.  

Example:
{quote}
message NodeHealthStatusProto {
    optional bool isHealthy = 1;
    optional string nodeHealthDescription = 2;
    optional string exceptionString = 3;
    optional NodeHealthDetail nodehealthDetail=4;
    optional StringIntMapProto nodeHealthdetail=5;
}

message StringStringMapProto {
  optional string key = 1;
  optional int32 value = 2;
}

keys could be - overall , ssd, non ssd, etc.. 
{quote}

Also make the NodeHealthService pluggable to support custom implementations of NodeHealthServices.

> Improve scheduling of containers based on node health
> -----------------------------------------------------
>
>                 Key: YARN-10335
>                 URL: https://issues.apache.org/jira/browse/YARN-10335
>             Project: Hadoop YARN
>          Issue Type: Improvement
>            Reporter: Bibin Chundatt
>            Assignee: Cyrus Jackson
>            Priority: Major
>
> YARN-7494 supports providing interface to choose nodeset for scheduler allocation.
> We could leverage the same to support allocation of containers based on node health value send from nodemanagers



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org