[ 
https://issues.apache.org/jira/browse/MESOS-5964?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Alexander Rukletsov updated MESOS-5964:
---------------------------------------
    Description: 
Currently, Mesos supports a single "restart policy" ("restart policy" is a poor 
name, alternatives are "unhealthy policy", "health policy"): kill a task when 
it becomes unhealthy. This can be described as "local health policy", because 
the decision is made locally.

In contrast, a framework may want to apply "global health policy", which means 
health status is reported to the scheduler and no action is taken on the agent; 
the scheduler decides how to act and instructs executor explicitly. A third 
policy can be "mixed": executor does kill tasks without consulting the 
scheduler, but it nevertheless propagates task health updates to the scheduler.

  was:Currently, Mesos supports a single "restart policy" ("restart policy" is 
a poor name, alternatives are "unhealthy policy", "health policy"): kill a task 
when it becomes unhealthy. This can be described as 


> Support health policies in health checks.
> -----------------------------------------
>
>                 Key: MESOS-5964
>                 URL: https://issues.apache.org/jira/browse/MESOS-5964
>             Project: Mesos
>          Issue Type: Improvement
>            Reporter: Alexander Rukletsov
>              Labels: health-check, mesosphere
>
> Currently, Mesos supports a single "restart policy" ("restart policy" is a 
> poor name, alternatives are "unhealthy policy", "health policy"): kill a task 
> when it becomes unhealthy. This can be described as "local health policy", 
> because the decision is made locally.
> In contrast, a framework may want to apply "global health policy", which 
> means health status is reported to the scheduler and no action is taken on 
> the agent; the scheduler decides how to act and instructs executor 
> explicitly. A third policy can be "mixed": executor does kill tasks without 
> consulting the scheduler, but it nevertheless propagates task health updates 
> to the scheduler.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to