Workflow level timeouts

[vicaire]

Is this a BUG REPORT or FEATURE REQUEST?: FEATURE REQUEST

The following issue proposes to add an exponential backoff retry strategy at the container level:

#700

Would it also make sense to provide retry strategies for whole workflows?

Additionally, would it be possible to have timeouts for both containers and complete workflow so that (independently of retries) the container and/or workflows are cancelled if they do not complete within a specified duration?

In some cases, it might be useful to specify different timeouts for the case where the execution of the container takes too long, or the scheduling of the container takes too long (due to a lack of ressources in the cluster).

[jessesuen]

At the container level timeouts already possible via the activeDeadlineSeconds flag which we carry forward to the pod.

At the workflow level, we have built the groundwork for supporting this (but not yet fully implemented) through the /execution annotation which all workflow pod executors will look to, when deciding if a pod should terminate. Currently the only user of this annotation is daemon steps, but it was intended to support all types steps, not just daemon. Workflow level timeouts would be implemented by setting the deadline in this annotation

[vicaire]

Sounds good. Thanks Jesse.

[vicaire]

Thanks!