If a task fails, it currently fails the whole query. It would be good to support some class of retries. Here's some ideas:
- retry tasks that have not emitted a row yet (ex. if a task is rejected due to load balancer)
- retry leaf tasks only (ex. so the retry only applies to an isolated part of the plan)
If a task fails, it currently fails the whole query. It would be good to support some class of retries. Here's some ideas: