add stop to experiment

[MarcusZuber]

Add a possibility to stop a running experiment without killing all tasks.

ToDo:

• Implementation in the base experiment
• Implementation in the different experiments
• Tests
• Documentation

[tfarago]

How is this different from calling ex.run().cancel()?

[MarcusZuber]

It's class method, so you can stop even when you forget to store the handle to exp.run() and it is more structurized stop that tries to finish stuff in a that you can actually re-run the experiment afterwards (e.g. if a cont. motor is moving the cancel would only stop it in the acceleration phase).

[tfarago]

I see, alright.

[MarcusZuber]

I thought to catch the CancelError and then make a structured stop, but then this would shadow the cancel of the motors etc. which might be also required in some cases.

[codecov]

Codecov Report

Merging #484 (3f6ef18) into master (66d9bb9) will increase coverage by 0.01%.
The diff coverage is 92.59%.

❗ Current head 3f6ef18 differs from pull request most recent head 2adc08a. Consider uploading reports for the commit 2adc08a to get more accurate results

@@            Coverage Diff             @@
##           master     #484      +/-   ##
==========================================
+ Coverage   87.53%   87.54%   +0.01%     
==========================================
  Files         114      114              
  Lines        7819     7843      +24     
==========================================
+ Hits         6844     6866      +22     
- Misses        975      977       +2     

Impacted Files	Coverage Δ
concert/tests/integration/test_experiments.py	99.51% <83.33%> (-0.49%)	⬇️
concert/experiments/imaging.py	92.28% <90.90%> (-0.09%)	⬇️
concert/experiments/base.py	89.06% <100.00%> (+0.53%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 66d9bb9...2adc08a. Read the comment docs.

[tfarago]

Not quite there yet I think.

[tfarago]

Also, instead of checking this in every experiment class, we can catch a stop request in the Acquistion, namely you can wrap the broadcast into something which accepts a stop request.

[MarcusZuber]

But what will it change? I have to write everywhere, where right now is if self._stop a handler of the stop request?

[tfarago]

You will have one check in the Acquisition class instead of N checks where N is the number of all possible producers.

I have to write everywhere, where right now is if self._stop a handler of the stop request?

Of course not, you don't have to write anything in the producers, that's the point.

[MarcusZuber]

But a clean stop is very experiment specific. So I abort all the acquisitions and then call a clean-up function, that has to take care that everything is back in a well defined state?

[tfarago]

In all your producers you just do if stop then break, so no clean up. And of course if special handler is needed you can always override the Experiment.stop method.

[MarcusZuber]

How about this? It is basically the cancel, but with a different exception. So in a generator I can catch it and make a structured stop.

[MarcusZuber]

Oh no:

asyncio.Task inherits from Future all of its APIs except Future.set_result() and Future.set_exception().