[Planetlab-users] Sirius Calendar Service Fixed

[David E. Eisenstat]

Hi all,

As the current operator of Sirius, I'm writing to tell you that Sirius is 
now functional again.

The problem that Robert Ricci discovered has been traced to a bad 
interaction between Node Manager and Sirius. When the slice to be boosted 
was not instantiated on a particular node, Sirius would blindly attempt to 
grant resources anyway. Node Manager in turn would leak these resources 
until it was rebooted by a daily cron job or otherwise, causing future 
attempts to grant resources to fail. This behavior has existed since 
mid-September at the latest, which is when I deployed a rewritten version 
of Sirius.

Sirius now checks for the presence of a slice before granting resources; 
NM will not be fixed until 4.0.

If you would like to verify the priority increase, Sirius attempts to 
contact your slice on a UNIX domain socket. Details here: 
http://snowball.cs.uga.edu/~dkl/faq.php (How will I know exactly when the 
priority increase occurs?)

I apologize for any inconvenience this may have caused you.

Best,

-David Eisenstat