On 02/24/11 - 09:51:32PM, William Henry wrote:
Hi
What is meant by "Condor must have "suspend"/stop/restart support."?
I think Condor supports these semantics so I want to understand it more.
In the condor deltacloud driver, one job == one instance. This may be a
mis-understanding on our part, but as far as we can tell the states that a
condor job can be in are:
IDLE (I)
RUNNING (R)
REMOVED (X)
COMPLETED (C)
HELD (H)
TRANSFERRING_OUTPUT (?)
Right now, a deltacloud job stays in IDLE from the time it is submitted up
until the time that the backend cloud actually launches the instance. At
that point, it goes to RUNNING. If the user decides to kill the backend job
through condor_rm, then the job goes into REMOVED and then COMPLETED. If
there is an error at any time, the job goes into HELD. We don't use
TRANSFERRING_OUTPUT.
When we want to support stateful instances, then when the user wants to "pause"
the instance, they need some way to tell condor to pause/suspend that job.
We don't currently see a clean way to do that. One idea is to have the user
put the job in HELD, and then later on use the "hold reason" to determine if
the job failed because of an error, or because we paused it. But it is
pretty ugly.
Do you have more details on how Condor supports suspend/stop/restart semantics?
--
Chris Lalancette