[
https://issues.apache.org/jira/browse/MYRIAD-105?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Santosh Marella reassigned MYRIAD-105:
--------------------------------------
Assignee: Santosh Marella
> Support NM placement
> --------------------
>
> Key: MYRIAD-105
> URL: https://issues.apache.org/jira/browse/MYRIAD-105
> Project: Myriad
> Issue Type: Improvement
> Reporter: Santosh Marella
> Assignee: Santosh Marella
>
> Today Myriad exposes a REST end point for operators to expand the YARN
> cluster by X number of Node Managers. The following is an example of the API:
> {code}
> PUT /api/cluster/flexup
> {"instances": 4, "profile": "medium"}
> {code}
> The above API doesn't impose restrictions on "where" the NMs would be
> launched. Thus, Myriad freely chooses to launch a NM on ANY slave the master
> has offered the resources on, as long as the offered resources are sufficient
> to launch the NM with the specified profile.
> However, for production deployments, the operators would like to have the
> flexibility to control/optimize placement of NMs on a subset of slaves. This
> is because, the operators have the best knowledge about the type of machines
> in their datacenter, the network topology for the machines, data locality
> etc.
> The two most common usecases seem to be:
> 1. Launch NMs on slaves matching certain hostnames/racks (for e.g.{code}
> "hostname LIKE prod-[1-12].example.com", "rack_id LIKE "rack-[1-3]"{code}).
> 2. Launch NMs on slaves with certain mesos slave attributes (for e.g.
> {code}"os LIKE centos6", "hdfsDataNode LIKE true"{code}).
> These are very similar to how marathon provides task placement via
> [constraints]
> (https://github.com/mesosphere/marathon/blob/master/docs/docs/constraints.md)
> in its REST API.
> Myriad can start small, perhaps with just LIKE operator.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
