[ 
https://issues.apache.org/jira/browse/MESOS-3155?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14646346#comment-14646346
 ] 

James Peach commented on MESOS-3155:
------------------------------------

The implementation we discussed was a (system) process that creates {{N}} 
libprocess processes, each of which registers as an independent slave. This 
would mean you can run a single command and generate {{N}} slaves (which 
publish an arbitrary resources). These null slaves don't need to actually run 
the tasks, they just need to correctly allocate resources. There are enough 
fields in the {{TaskInfo}} that you can designate specific triggers for how the 
virtual task should behave (e.g. you could use the {{Environment}} for this). 
The null slaves can introspect this information to make tasks fail, succeed 
after specific timeouts, etc.

In principle, we would like to be able to use the null slave with our internal 
scheduler, but we haven't figured out how to deal with the task image in that 
context yet. Maybe it won't turn out to be much of a problem.

> Add a containerizer and executor that simulate the launching of tasks. 
> -----------------------------------------------------------------------
>
>                 Key: MESOS-3155
>                 URL: https://issues.apache.org/jira/browse/MESOS-3155
>             Project: Mesos
>          Issue Type: Improvement
>          Components: containerization, slave
>            Reporter: Cody Roseborough
>            Assignee: Cody Roseborough
>            Priority: Minor
>
> The Simulator Containerizer (SC) is a containerizer that can be used in 
> conjunction with mesos-local to exercise frameworks at a large scale without 
> launching a large cluster. The SC launches a "fake" executor instead of the 
> one specified by the tasks, in order to avoid using actual resources. The SC 
> provides normal tasks updates as well as configurable task options via a 
> configuration JSON file. 
> Some of the options available for configuration might be length or failure 
> rate. 
> Other points of contact for this are: [~lukeleslie], [~derekchiang]



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

Reply via email to