New feature: retry decorator for specific exceptions

[dmfigol]

Task queues like celery, have a built-in decorator to handle retry for specific exceptions:
http://docs.celeryproject.org/en/latest/userguide/tasks.html#retrying
I think we could implement something similar, where you could specify a list of exceptions for the task for which you want to retry and how many times you want to retry.

[dbarrosop]

Not so sure about this, IMO this is much better solved by python:

def my_task_with_retries(task, retry=0):
      try:
            r = task.run(...)
      except MyException:
            if retry < MAX_RETRIES:
                 r = my_task_with_retries(task, retry+1)
            else:
                 raise
      except SomeOtherException:
            r = some_other_task(task)
      finally:
            cleaning_operations(task)
      return r

With a decorator this level of flexibility is hard to achieve and ends up polluting the core (which I think it should be as simple and lean as possible to minimize the chances of bugs).

Also, note that a grouped_task is already a sort of decorator where you can wrap any task plugin and implement your own workflow.

[dbarrosop]

This is similar to the discussion we had (offline, sorry about that) here: #3

[dmfigol]

@dbarrosop currently library exception is not raised, only NornirExecutionError. This prevents a user from doing what you showed above. One would need to use failed_hosts and result.exception to make something like this work.

[dbarrosop]

We could easily reraise the original exception in the subtask case. Either that or just something like:

try:
...
except NornirSubTaskError as e:
    if isinstance(e, AnException):
        ...
    elif isintance(e, SomeOtherExeption):
        ...

[dmfigol]

Could you give the whole example with let's say some Netmiko exception or Napalm?