#344: Checkpointing + new cleanup procedure --------------------------+------------------------------------------------- Reporter: mthomas | Owner: mthomas Type: enhancement | Status: new Priority: blocker | Milestone: Component: SimFactory | Version: Resolution: | Keywords: --------------------------+-------------------------------------------------
Comment (by hinder):
I have been using the previous patch which only fixed up checkpointing, so I can only comment on that. I haven't tried this one. With the previous patch, recovery works if the job is not currently in the queue. If the job *is* in the queue, and is running, the new job is chained to the old one. If the job in the queue is only in the Q state, however, it does not get detected and the new job is also put into the Q state, not the H state. It might be that the logic for whether to chain or queue the job depends on whether it has run or not, which is not correct, I think.
Barry: can you check if the following four cases work correctly with this patch?
1. If no jobs for that simulation are in the queue; 2. If a job is in the queue in the Q state; 3. If a job is in the queue in the R state; 4. If a job is in the queue in the H state (should only be the case if 2. is true as well, but best to check)
My own tests are on Kraken, which can have long turn-around times, so you might want to test somewhere else like Damiana or Datura. If it works on those machines, then the code is probably OK, and at worst the Kraken machine entry might need to be fixed up.