mark_completed() causes kernel to die

[gorogm]

Thank you for helping us making ClearML better!

Describe the bug

task.mark_completed() causes the kernel (in .ipynb and .py without as well) to die task.close() works better - kernel doesn't die

To reproduce

import matplotlib.pyplot as plt
import numpy as np
from clearml import Task
import socket

task = Task.init(project_name='Szemet', task_name='loo clean')
task.set_parameter('hostname', socket.gethostname())
plt.plot(np.random.random(10))
plt.show()

task.mark_completed()

Expected behaviour

Kernel should continue running, with task marked as completed

Environment

• Server type - app.clear.ml
• ClearML SDK Version 1.6.4
• Python Version - 3.9.12
• OS (Windows \ Linux \ Macos) - Ubuntu 22.04

Related Discussion

Similar behaviour was reported by others: https://clearml.slack.com/archives/CTK20V944/p1655149087316649 and https://clearml.slack.com/archives/CTK20V944/p1653505498243069

[erezalg]

Hi @gorogm,

Yup :smile: that indeed seems to happen and it's not supposed to.

We'll make sure to fix it

[erezalg]

Hi @gorogm,

I had a chat with our developers and it actually is by design that this happens. basically, task.mark_completed() is meant for scripts that manage other tasks (such as a hyperparameter optimization controller or a pipeline controller). The correct method to use is task.close() to close the currently running task. We will update the docs to reflect that usage, but as it stands, please use task.close()

Makes sense?

[gorogm]

Very nice, thanks a lot!! :)