Add on_error hook in pipeline definition to be executed when a task errored

[hlubek]

Scenario: A task like a database migration needs a rollback if it errored

[Benjamin-K]

on_error could also be used, if a job is cancelled. Or do you think, this should be treated differently?

[hlubek]

@Benjamin-K If the job was already started - and then canceled - it would make sense. Question is, if the name on_error is suitable for both the error case and rollback on cancellation and the behaviour could be expected on cancellation.

[Benjamin-K]

@hlubek What if we add some variables to the on_error hook? Sth. like PRUNNER_PIPELINE_STARTED=1, PRUNNER_FINISHED_TASKS=task1,task2,task3 (or as a list) so one could access them to decide, whether to do some rollback or not?