Github user markhamstra commented on the pull request:
https://github.com/apache/spark/pull/186#issuecomment-38337656
Okay, you're persuading me. I was thinking about exceptions being
generated only by misbehaving jobs whose ill effects would only be present in
the DAGScheduler. Failures in the DAGScheduler itself we can at least hope to
recover from if we can do sufficient cleanup of its state in the restart hooks,
but if the exceptions are caused elsewhere in Spark, just surfacing first in
the DAGScheduler, and require cleanup/restart actions outside of the
DAGScheduler's control, then I think you are right, Kay, that even more extreme
actions are called for.
In the future, we may want to look at whether we can cleanly segregate
exceptions caught in the DAGScheduler into different levels of recoverability
so we don't have to default to taking extreme actions that will kill the
SparkContext, kill multiple jobs, etc. But that's definitely not something
that we can do for 0.9.1.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
