Bug 6631

Summary: inconsistency with resource reservations
Product: Nimbus Reporter: Tim Freeman <tfreeman@mcs.anl.gov>
Component: Workspace serviceAssignee: Tim Freeman <tfreeman@mcs.anl.gov>
Status: RESOLVED FIXED    
Severity: normal CC: keahey@mcs.anl.gov, workspace-dev@globus.org
Priority: P3    
Version: TP2.2   
Target Milestone: TP2.2.1   
Hardware: PC   
OS: Linux   

Description From 2009-01-27 15:02:25
It is possible to get into an inconsistent state with the resource
reservations.

Conditions:

1. Cluster request is granted all its network requests
2. Ensemble-done called (actually request cluster)
3. VMM slots are partially granted but there are not enough to satisfy the
whole cluster.
4. Backout is invoked

The logical error is in the code for step #4, in
DefaultSlotManagement#reserveCoscheduledSpace()

This was typically never exercised in the past because most deployments have
less or equal network addresses than VMMs (i.e., requests usually don't get by
step #1 if they want public addresses).
------- Comment #1 From 2009-01-27 15:08:43 -------
committed fix