Discussion on Pipeline breakpoints feature

[davidsbatista]

This is the discussion board for the experimental feature of adding breakpoints to a Pipeline

This experiment allows set breakpoints at any component in any pipeline, forcing the pipeline run to stop before that component runs.

Usage Examples

# Setting breakpoints
pipeline.run(
    data={"input": "value"},
    breakpoints={("component_name", 0)},  # Break at the first visit
    debug_path="debug_states/"
)

This will generate a JSON with the complete pipeline state before the next component is run, i.e.: the one receiving the output of the component set in the breakpoint

# Resuming from a saved state
state = Pipeline.load_state("debug_states/component_state.json")
pipeline.run(
    data={"input": "value"},
    resume_state=state
)

🧑‍🍳 See an example notebook here
💻 See the PR that adds the experiment here

Note: Experimental features live in this repository for a fixed period of time. We don't guarantee that we will continue maintaining experimental features. But if they are successful, stable and if you like it, we will move the feature to the core Haystack package.

[aymbot]

Great feature! As a developer working with haystack, I would like to know if this feature will be expanded to async pipelines as well?
I am in the process of developing a multi-agent architecture where some of the agents are executed within async pipelines, hence the interest.

[davidsbatista]

Yes, we are also planning to have it in the async version of the Pipeline. Can you test it as well on the sync version, or you can only test it in the async?

[sky-2002]

@davidsbatista this would be an interesting feature
I was thinking that this could also be used for cases where in a workflow, if a certain component fails, we can just save the state and then it picks up from where it left in a later try.
Would like to know your opinion on how breakpoints could be used in such cases, where we don't need to run the earlier components again, just because some component failed.

[davidsbatista]

Hi @sky-2002! That's an interesting use case.

Currently, you always need to set up breakpoints and associate them with a component. The pipeline will then interrupt its execution before a component associated with a breakpoint is executed.

What you suggest is something else, an interesting approach to allow for debugging/resuming a Pipeline when it breaks/crashes. The use case you describe would need some changes in the pipeline code, but one could reuse the structure of the Breakpoint.

We need to move this discussion to haystack main, since the breakpoints were already merged into main.

In any case, I encourage you to open an issue with a more details about the suggestion you did, we can then pick it up and refine it.

[davidsbatista]

Pipeline and Agent breakpoints were released with haystack 2.16.0. Please use the discussion board from haystack main for any comments/suggestions regarding the Pipeline breakpoints.