add support to automatic retry of tasks

[kaazoo]

Like in the old releases, it should be possible to specifiy the number of retries when creating a new task.
In load balanced view, if one task fails on one engine that task could be rerun on another engine.

minrk suggests the following workaround, which makes a copy of the same task depend on the original one and running that only if the first try fails.

ars = [view.apply_async(task)]
with view.temp_flags(after=Dependency(ars, success=False,failure=True)):
    ars.append(view.apply_async(task))

This workaround might be ok if you only want to have 1 retry. But if you need a higher number of retries you will create lots of tasks which might never be executed.

[minrk]

Can you test from my retries branch: #413?

It adds retries as an integer flag, specifying the maximum number of retries.

I believe that should behave in the same way as the previous version.

[kaazoo]

Thank you. Setting the retry limit works. If I create tasks that will fail (for example try to render a Blender scene file which isn't existing) leads to having them atuomatically requeued to other engines.