oscarsaleta / pbala Goto Github PK

If fork fails, the slave notifies the master, but the master is programmed to send a job to the slave and does not listen to possible errors when forking. Thus, we can have a deadlock (both pieces trying to send messages to the other one but none listening).

The result is that the auxprog files are not removed (at least for Sage executions), leaving the results directory cluttered.

The execution itself works correctly.

When killing Sage jobs (e.g. killing the GP processes, or the Python processes), the program returns to Sage, and Sage exits without the killed flag, so the job is not registered in the unfinished_tasks.txt file.

PBala doesn't give tasks to a node if it has too low memory, but it also fails to give the task to another node, making executions fail if a node in nodelist has no memory left because of another user.

The user might not be able to get information about the execution until it is completed if they redirect stderr to a file.

I've added setvbuf(stderr,NULL,_IOLBF,BUFFER_SIZE) to the code but doesn't look like it's working

From X to taskArgs

Would be... Total time of task being alive, or actual execution time by task? Could do:

• a time() when tasks are spawned and another when each task is killed,
• ime each execution inside the tasks and return the sum to the master.

This is a warning by gcc because a variable of type struct rusage is used as argument of a function defined in the program, but gcc fails to notice that is not the declaration of such type.

Required libraries are included and the program works, so the warning does not affect the execution.

How to get rid of it?