Button on UI to pause all DAGs

[yshkp]

Description

Slider button on UI of airflow to pause/unpause all DAGs.

Use case/motivation

During deployment, every time there is a use case to pause all dags to not run tasks in the old build. So this button will come handy during deployment.

Related issues

No response

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[boring-cyborg[bot]]

Thanks for opening your first issue here! Be sure to follow the issue template! If you are willing to raise PR to address this issue please do so, no need to wait for approval.

[coolbeans201]

If there's a button for this in the UI, we'd definitely want to make sure that only the highest level of roles within a given server have access to this feature. That's probably assumed and would be taken care of, but just want to state the obvious.

[cnkumar20]

this is not ideal if the UI button doesn't revert dags back to its original state on unpause. having list of dags as params should be strictly required for this for pause /unpause to original state. should be covered in the caveat

[rsdel2007]

Hey @potiuk I am willing to work on this.

[potiuk]

Cool

[eladkal]

I would rather this not be a UI feature. This is very dangerous operation and with a single click of a button that was done by mistake can cause serious problems to large deployments.

If this must be a UI button it must come with seperated role with default of usage only to Admin. However I would rather this functionality to be API only. To my prespective (and correct me if I am wrong) this operation should be used rarely so having it as button in the main page might be not the best?

WDYT?

[SamWheating]

Agreed that this is dangerous as the implied "unpause all" can resume DAGs which were previously paused for good reason.

As an admin, I have the option to scale the scheduler down to 0 replicas (preventing any runs from advancing) or setting AIRFLOW__SCHEDULER__USE_JOB_SCHEDULE=False (preventing any new DAG Runs from being created). So far this has been sufficient, but maybe there's a use-case I'm missing here.

[cmarteepants]

What if we switched from a toggle per dag to a multi-selector with a pause/unpause action button? Similar to what we have here:

It's no longer one-click so it would be harder to accidentally pause all dags, and you can pause a subset more easily if you want. However, it does mean more clicks if you're only pausing one dag.

FWIW, I think the env var is fine for the limited times you need to do this but I can appreciate that people like buttons. I may also missing a use-case.

[rsdel2007]

@potiuk I agree with @cmarteepants suggestion. This seems to be the right way. WDYT?

[potiuk]

I am fine with that. @eladkal ?

[eladkal]

What if we switched from a toggle per dag to a multi-selector with a pause/unpause action button? Similar to what we have here:

Can you be specific into what view you are referring to?
You mentioned similar to the task instance view can you explain which view you had in mind? The tasks instance and the dag run views are not suitable for this button.

[cmarteepants]

I agree - the task instance and dag run view are not the right the place - it was the only view I could find that had a multi-selector. My idea was to replace the individual toggles to pause/unpause dags with a multi-selector + action button in the home page / dag list view.

[cmarteepants]

Thinking about this further: Another idea is an extension of @yshkp's original proposal, which takes into account whether individual dags were paused before globally pausing all dags. We proceed with a button to pause/unpause all dags, however:

• When dags are paused, we throw a warning box to confirm that this is intentional
• We add a banner saying that all dags are paused
• We disable all the pause/unpause toggles, but leave them in their current position. This would act as a visual identifier to see whether a dag was paused before someone globally paused all dags
• Dags have to be unpaused via the same button.
• When dags are unpaused, we enable all the toggles and dags will revert back to their individual paused/unpaused states

To be fair, this is becoming a lot more complicated.

[eladkal]

To be fair, this is becoming a lot more complicated

I belive its complicated because you are combining the business logic of the dag with adminstative operation of upgrade. DAG author should not have business logic impacted (on/off) because cluster admin decided to make an upgrade. Meaning that if you switch my dag to off for upgrade then you need to turn it on once upgrade is done how would you do that for a deployemnt of 1000 dags authored by 200 different developers?

I think you are consedring only deployments where the dag author and the cluster admin are the same person :)
This is why I say it is a dangerous operation and since the use case you described is for upgrade I suggested to not make it a UI button but API call.

I am happy to discuss ideas of how to accommodate your request but you need to understand that we have to consider all kinds of deployments and the implication of such button would make.

I would be more happy with adding such operation on a folder base rather than all dags once #24464 is implemented.

(On side note I still don't quite follow why termination of scheduler is not good enough to handle your use case)

[cmarteepants]

I'll defer to others on the use-cases. These were just two drive-by suggestions on how we can add UI capability for this while minimizing risk, assuming this is something the community truly wants :) Personally, I think the existing env var solution is fine.