2012年1月10日 下午 02:01
we have quite a simple set up here on our HPC Server 2008 R2: one cluster admin ('administrator') and one widely used cluster user account ('fred'). Until yesterday, fred was also a cluster admin by the HPC Cluster Manager. Fred ist also a member of the Administrators, Domain Admins, Domain Users and HPCUsers group on the head node. The head node is the domain controller for a separate hpc domain.
Everything was fine until we decided that the normal users did not need the HPC Admin privileges and set the role of fred to User in the Cluster Manager. With that, fred cannot use the Job Manager any more (everything greyed out), but he can submit jobs via a different application, but those jobs fail because fred is suddenly no longer allowed to start or stop services on the compute nodes (which we do as node prep and release tasks for our specific jobs).
What's wrong? Why do we need cluster admin privileges?
2012年1月10日 下午 05:40
Thank you for the question, the simple answer is that when you job executes on a compute node, it has the same privilege as the user initiating the job.
Note that only admin can start/stop services, so when Fred initiates the job, the job runs as a normal user on the compute nodes as Fred, then fails because Fred cannot start/stop services.
Can you double check the privilege for Fred on your compute nodes? I believe your problem will be resolved when Fred is in the admin group of your compute nodes.
Hope it helps.