-
Error checking in ssh, pbs and fork engines, allowing to check when a worker has exited without computing its result. Once a worker fails, it is possible to repeat the task for a limited number of times before marking it broken and throwing an error.
-
Add cancel operation to future or engine objects, in such a way that any job can be aborted using this method.
-
Yield support for individual tasks, as ones executed by "px.run" function, in such a way workers can yield results to the server and the server can give parameters to the worker through join function in its corresponding future object.
-
Use time field in future objects to control timeout or connections problems.
-
In qsub engine, cancel all pending/executing jobs in PBS when exiting the program.
-
Keep a checkpoint list at user folder which can be useful to recover from failures. To protect this checkpoint list the file will be keeped at ~/.parxe/checkpoints instead of tmp folder.