-
Is it worth to introduce automatic serialization of the Task? For pure python tasks, this would remove the need for a command-line interface, i would instead deserialize and run. For non-python jobs, i would deserialize, create the runner script from the template and run it using the subprocess module.
-
? How many jobids can i use as dependencies (-hold_jid/-d)? Are there any limits?
-
? Can i use array jobs
- some templating + job args in a file + line per task(?)
- even easier just create a file with all the arguments per task - task-$ARRAY_INDEX.args
-
? How to make chunking work with array jobs?
-
Keep a db of submitted jobs
- Is this worth the mess? Would it allow restarts?