Removing all auxiliaries interfaces to only keep double-threaded interface

[sebastianpfischer]

Status	OPEN
Participants
Outcome	Option 1: we delete it
Decided On	21.07.2023
Due Date	30.08.2023
Owner	@sebastianpfischer

Problem Description

Due to performance issues, we had implemented many different possible interfaces and also used them.
When we got more time, we basically designed and @BKaDamien implemented it, a so called flexible double-threaded-auxiliary interface.

This enabled us to cover with one configurable interface all the other ones we had. Therefor, after migrating all auxiliaries to the new interface, the next step would be to remove all the previous interfaces (would improve our maintainability)

There is one Use-case / interface we nevertheless do not really cover, it is the multi-processing interface one. Today, pykiso, due to many performance improvements, do not rely on it anymore but what if, in the future we would need it?

Possible use-cases

• image processing based auxiliary. Use a model to detect something on an image flux.
  • handling the stream uses CPU
  • running the algorithm on it may take a lot of CPU
• data-intensive algorithm with ML

Influencing Factors

• Our quality goals:
  • Modularity - "The framework shall allow me to implement complex logic and to run it over any communication port"
  • Portability - "The framework shall run on linux, windows, macOS"
  • Performance - "The framework shall use only the right/reasonable amount of resources to run (real-time timings)"
• Does it scale cross multiple core?
• To add more next meeting.

Assumptions

Is there anything we need to assume about the problem?

Alternatives

1. remove everything
2. leave the way to create an own aux interface
3. leave the multi-processing interface and proxy available
4. Create a special aux example showing how to port intensive work into other cores.

Decision Matrix

Until now, we could find a use-case strong enough to justify any maintenance of the multi-processing API. Therefor we take the risk to remove it and later on to add it again if needed. The last release containing multiprocessing will be 0.23.0.

Final Decision

Option 1