mesos-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Joseph Wu (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (MESOS-7155) Mesos Maintenance Primitives ("Drain Mode" could not see it working)
Date Wed, 22 Feb 2017 16:34:44 GMT

    [ https://issues.apache.org/jira/browse/MESOS-7155?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15878648#comment-15878648
] 

Joseph Wu commented on MESOS-7155:
----------------------------------

All frameworks need to support maintenance primitives for "draining" to work properly.  That
means each framework must use the V1 HTTP API (instead of the c++ native driver, which does
not receive inverse offers) and each framework must handle the {{INVERSE_OFFERS}} event.

> Mesos Maintenance Primitives  ("Drain Mode"  could not see it working) 
> -----------------------------------------------------------------------
>
>                 Key: MESOS-7155
>                 URL: https://issues.apache.org/jira/browse/MESOS-7155
>             Project: Mesos
>          Issue Type: Documentation
>          Components: agent, documentation
>    Affects Versions: 1.0.0
>            Reporter: Rahul Bhardwaj
>            Priority: Critical
>              Labels: features
>
> Hi,
> We are going by Mesos Maintenance Primitives documentation here http://mesos.apache.org/documentation/latest/maintenance/.
My requirement is "that during a Agent maintenance we want to move all running tasks from
this agent to other Agents without task failed/stop". This is how you do zero downtime maintenance
without affecting the running tasks.  I see that in the documentation  "we submit a maintenance
schedule" which sends inverse offer to Framework to plan tasks according to the agent(s) unavailability.
But Drain mode is not working here. I mean b/w  submissions maintenance-schedule and Agent-down,
tasks are not moved from the agents going under maintenance to other agents. As a result we
cannot achieve a 100% full proof maintenance process. 
> Can you please elaborate on the "framework respond to inverse offer" process or "Drain
mode" step. This is very critical part in the maintenance. Other steps are fine and working
for us (de-registering and re-registering agent from&to cluster)
> Thanks



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message