[
https://issues.apache.org/jira/browse/MESOS-3044?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14631921#comment-14631921
]
Joseph Wu commented on MESOS-3044:
----------------------------------
Note about the "Won't Fix" resolution: We decided that there are too many edge
cases for the master to safely deactivate a slave once the given maintenance
window passes. Instead, all maintenance-related mode changes must be initiated
by an operator. See the design document for more details.
> Slaves are not deactivated upon reaching a maintenance window
> -------------------------------------------------------------
>
> Key: MESOS-3044
> URL: https://issues.apache.org/jira/browse/MESOS-3044
> Project: Mesos
> Issue Type: Task
> Components: master
> Reporter: Joseph Wu
> Assignee: Artem Harutyunyan
> Labels: mesosphere
>
> After a maintenance window is reached, the slave should be deactivated to
> prevent further tasks from utilizing it.
> * For slaves that have completely drained, simply deactivate the slave. See
> Master::deactivate(Slave*).
> * For tasks which have not explicitly declined the InverseOffers (i.e.
> they've accepted them or do not understand InverseOffers), send kill signals.
> See Master::killTask
> * If a slave has tasks that have declined the InverseOffers, do not
> deactivate the slave.
> Possible test(s):
> * SlaveDrainedTest
> ** Start master, slave.
> ** Set maintenance to now.
> ** Check that slave gets deactivated
> * InverseOfferAgnosticTest
> ** Start master, slave, framework.
> ** Have a task run on the slave (ignores InverseOffers).
> ** Set maintenance to now.
> ** Check that task gets killed.
> ** Check that slave gets deactivated.
> * InverseOfferAcceptanceTest
> ** Start master, slave, framework.
> ** Run a task on the slave.
> ** Set maintenance to future.
> ** Have task accept InverseOffer.
> ** Check task gets killed, slave gets deactivated.
> * InverseOfferDeclinedTest
> ** Start master, slave, framework.
> ** Run task on slave.
> ** Set maintenance to future.
> ** Have task decline maintenance with reason.
> ** Check task lives, slave still active.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
