Timing bug (RuntimeError) in running analysis with block_for_results()

[itoko]

Informations

• Qiskit Experiments version: 6a732f4
• Python version: 3.10.10
• Operating system: Mac OS

What is the current behavior?

Running an analysis (e.g. Parallel T1) with block_for_results() occasionally fails with a RuntimeError: cannot schedule new futures after interpreter shutdown. It may suggest block_for_results() does not actually wait for the completion of results.

Steps to reproduce the problem

Running the code below via command line.

import numpy as np
from qiskit_experiments.framework import ExperimentData, ParallelExperiment
from qiskit_experiments.library import T1
from qiskit_ibm_provider import IBMProvider

# run and save parallel experiments in advance
provider = IBMProvider()
backend = provider.get_backend("ibm_wellington")
exps = [
    T1(
        physical_qubits=[q],
        delays=np.arange(1e-6, 6e-4, 5e-5),
    )
    for q in range(backend.configuration().num_qubits)
]
exp = ParallelExperiment(exps)
exp_data = exp.run(backend, shots=1000)
exp_data.auto_save = True
exp_data.block_for_results()

# load and rerun analysis
new_exp_data = ExperimentData.load(exp_data.experiment_id, provider=provider)
new_exp = ParallelExperiment(exps)
new_exp.analysis.run(new_exp_data)
new_exp_data.block_for_results()

Try rerun the "load and rerun analysis" part with your experiment_id if no error happens in your env.

What is the expected behavior?

block_for_results should wait for the completion of analysis.

[gadial]

The reason for this behavior (which is not a bug) is that the line new_exp.analysis.run(new_exp_data) creates a copy of new_exp_data (since run was called with replace_results=False. The solution is to do new_exp_data = new_exp.analysis.run(new_exp_data) instead.

This raises the question of whether we should use the default replace_results=False or change the default to True to avoid similar confusions.

[itoko]

@gadial Thank you for letting me know the spec. I should have read the API doc carefully.
I now understand this is an issue on this code in how-to guide, which I referred when writing the above code. It should be

expdata = experiment.analysis.run(expdata)

or

experiment.analysis.run(expdata, replace_results=True)

[coruscating]

Thanks @itoko and @gadial, I've updated the how-to here: #1339