User Tools

Site Tools


use-cases:node-deletion
Use case name

#11: Node deletion

Goal

Delete an existing node of the platform

Scope

Server (Software) and Nodes (Software)

Level

Summary

Pre-conditions
  1. The node is announced or registered on the sistem
  2. The user that performs the deletion (actioner) has to be registered
Primary actor

A registered actioner

Trigger

A registered user with some credentials level uses the server interface (or server API) to delete a node from other user (or himself)

Main Success Scenario
  1. The actioner selects a node and clicks on delete option
  2. As alternative: the actioner sends a valid request with all information to the Public server API
  3. The server verifies the actioner credentials
  4. The server verifies that the actioner current credentials are higher than the node's owner to delete
  5. The server verifies if the node actually is part of any sliver and/or if it's running an experiment
  6. The server notifies to each sliver that the node will be deleted
    1. LEANDRO: What is a notification? Who/how is notified?
  7. The server notifies to each user with an experiment on the node that the node will be deleted
  8. The server stops the node and removes all assigned slivers
  9. The server deletes the node from their database (if the node restarts, it won't be announced)
    1. LEANDRO: maybe not completely as if a node is deleted, that state should be remembered to avoid a node is re-registered immediately (so called 'death certificate').
Details

DAVIDE: Open topics are: If node is working or is part of a sliver, what happens?

IVAN: The owner should be informed of that, but we can't keep the owner from pulling the plug and crashing the slivers ;) . Of course that shouldn't be welcome, there should be a way for the node admin to state that no more slices are allowed (this doesn't solve the problem of long running slices).

LEANDRO: Not 100% sure but deletion and other changes relate to states of nodes (and actions that cause state transitions). We can check the node states in Planetlab (reinstall, boot, …) we could have states/actions such as: throttle (node not accepting new connections/slivers), down, restart, reinstall, shutdown, remove, suspected (not recent contact/status), running. A bit confusing as state can be the result of monitoring or actions/events originated at the server, the node or a separate monitoring service.

use-cases/node-deletion.txt · Last modified: 2012/04/15 22:21 by leandro