Github user brkyvz commented on the issue:
https://github.com/apache/spark/pull/17543
Yes, your explanation is on point. If I have 4+ executors that died, then
all retries of Stage B will also eventually fail. If we didn't ignore these
failures, we could have re-computed the outputs of Stage A for more of the lost
executors earlier on.
So, I'm also worried about the case you specified where executor Z was on
the same machine as executor X, and we don't necessarily have to invalidate.
That seems like an optimization to me, which actually leads to failures in
corner cases like mine. Maybe it's better to be safe than sorry?
I'll definitely add a unit test, I just want to make sure I'm solving the
problem the right way in the right place first :)
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
