action #64667
openSTART_DIRECTLY_AFTER_TEST - Child cancellation cancels other childs
0%
Description
problem / current behavior¶
In a context of START_DIRECTLY_AFTER_TEST, if a child is cancelled, then other childs are skipped.
Example:
B and C have START_DIRECTLY_AFTER_TEST=A
---> B
/
A
\
---> CA ran successfully, B is running and C is waiting that B is done to start.
If I cancel B, then C will not be run and is skipped.
expected behavior¶
As B and C are independents/siblings, cancelling B should start C.
Updated by mkittler over 4 years ago
- Description updated (diff)
- Target version set to future
In the chat you've been talking about priorities as well. So what you've observed has nothing to do with job priorities after all?
But I get the problem. That's indeed something I haven't really put thought into when implementing START_DIRECTLY_AFTER_TEST. I suppose what you're asking for is actually just a tiny adjustment in the worker code. Only testing will be annoying and likely only be possible manually.
I was so free to amend the description so I later still know what the current behavior is and what the desired behavior is.
Updated by ggardet_arm over 4 years ago
mkittler wrote:
In the chat you've been talking about priorities as well. So what you've observed has nothing to do with job priorities after all?
Job priority inside START_DIRECTLY_AFTER_TEST is a separate problem. Except if you want to fix as part of this ticket?
Updated by okurz over 4 years ago
- Category set to Feature requests
I guess different people can have different expectations, e.g. "if any child is cancelled in a cluster, invalidate the whole cluster", i.e. cancel other siblings as well. What's the use case behind cancelling B but keeping C?
Updated by ggardet_arm over 4 years ago
okurz wrote:
What's the use case behind cancelling B but keeping C?
B can be known broken, but we still want to test C which is a complete different test. The only commonality between B and C is they requires A to be run before.