Thanks for your suggestion @<1523701087100473344:profile|SuccessfulKoala55> !
It sounds like retry is hard if you're not using POST. Am I interpreting this right?
POST is convenient, yeah I agree but I don't see why the rest can't retry.

It would be helpful if you could elaborate more; which behavior you're trying to control, what's the actual SW stacks and why they're on good terms with POST, why other methods won't work, etc.

Thanks @<1523701070390366208:profile|CostlyOstrich36> , so i think some control logic must be written the code, and then use the pipeline input to select the portion of tasks to run (pls correct me if i'm wrong)

Then, i assume selectively running only some tasks from the web app is not supported yet? Let's say i want to start running the [ClearML example pipeline](https://app.clear.ml/pipelines/3572715422f15594b676fa9004e36149/experiments/e-7d38ec49f0cc41b99f67929474af994a?columns=selected...

Yeah all clear to me. Thank you.

the exported task

I think that what you want is to build a pipeline from those Tasks? Or create a Task and enqueue it manually directly from Task A?

Hi @<1523701205467926528:profile|AgitatedDove14> , exactly what I wanted 🙂 .
I was wondering about what i can do with the agent's argparse magic.
Thanks for your answer.

nvm, it's working somehow 🎉 . thanks a lot!

I don't think docker is there since the machine image field in the autoscaler config is limited to the public ones.

Will adding some init script to install docker do the trick?