You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@reef.apache.org by "Matteo Interlandi (JIRA)" <ji...@apache.org> on 2019/01/04 23:57:00 UTC
[jira] [Created] (REEF-2054) Elastic Broadcast
Matteo Interlandi created REEF-2054:
---------------------------------------
Summary: Elastic Broadcast
Key: REEF-2054
URL: https://issues.apache.org/jira/browse/REEF-2054
Project: REEF
Issue Type: Sub-task
Reporter: Matteo Interlandi
Assignee: Matteo Interlandi
This subtask is used to track the implementation of the elastic broadcast operator.
To have the elastic broadcast operator working the following components are required:
Driver side:
* a context implementation providing the entry point to the elastic service
* a stage implementation wrapping pipelines of +operators+ into logical schedulable units
* a task set object managing the scheduling and communication of tasks
Task side:
* a context implementation providing the entrypoint for tasks to the elastic service
* a stage implementation wrapping the +workflow+ of operators
* a workflow object surfacing the list of operators to the task and enabling to manager failures events by the elastic framework
* a communication layer between tasks and between tasks and the driver
Operators:
* for now only the broadcast operator will be implemented
* each operator will have:
** a set user defined policies in case failures occur
** a set of mechanisms implementing the policies
** a topology mirroring the logical organization of nodes in the network. For the moment only a flat topology (1 master, n nodes) will be implemented
** user definedĀ +failure machine+ defining when policies are triggered
** policies, mechanism and failure machines are used also at stage and context level (in a sort of hierarchical way) to provide high level behaviors under failures.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)