[jira] [Updated] (MESOS-2076) Implement maintenance primitives in the Master.

["Joseph Wu (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/MESOS-2076?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Joseph Wu updated MESOS-2076:
-----------------------------
    Description: 
The master will need to do a number of things to implement the maintenance primitives:

# For machines that have a maintenance window:
#* Disambiguate machines to agents.
#* For unused resources, offers must be augmented with an Unavailability.
#* For used resources, inverse offers must be sent.
# For inverse offers:
#* Filter them before sending them again.
#* For declined inverse offers, do something with the reason (store or log).
# Recover the maintenance information upon failover.

Note: Some amount of this logic will need to be placed in the allocator.

  was:
The master will need to do a number of things to implement the maintenance primitives:

# For slaves that have a maintenance window:
#* For unused resources, offers must be augmented with an Unavailability.
#* For used resources, inverse offers must be sent.
# For inverse offers that are declined, we must filter these before sending them again. We must also store the decline reason, guard against OOMing. 
#* My hunch is that we'll not want to persist the reasons in the initial approach.
# When the drain window is reached, we'll make a binary decision as to whether the slave was drained, based on whether it was empty.
#* If drained, we deactivate this slave and store the fact that it was drained.
#* If not drained, we leave this slave activated.
# Recover the maintenance information upon failover.


> Implement maintenance primitives in the Master.
> -----------------------------------------------
>
>                 Key: MESOS-2076
>                 URL: https://issues.apache.org/jira/browse/MESOS-2076
>             Project: Mesos
>          Issue Type: Task
>          Components: master
>            Reporter: Benjamin Mahler
>            Assignee: Artem Harutyunyan
>              Labels: mesosphere, twitter
>
> The master will need to do a number of things to implement the maintenance primitives:
> # For machines that have a maintenance window:
> #* Disambiguate machines to agents.
> #* For unused resources, offers must be augmented with an Unavailability.
> #* For used resources, inverse offers must be sent.
> # For inverse offers:
> #* Filter them before sending them again.
> #* For declined inverse offers, do something with the reason (store or log).
> # Recover the maintenance information upon failover.
> Note: Some amount of this logic will need to be placed in the allocator.



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