Github user squito commented on the issue:
https://github.com/apache/spark/pull/17113
> then you get fetch failure again and iterate until job failure
At first I was thinking the node goes bad, but you first detect it via
fetch failures -- in that case, you wouldn't need this, when the tasks get
re-run on the bad node, they'd fail, and the node would still get blacklisted.
But you're saying the issue is if the node is actually totally healthy, so
tasks can still write shuffle data, but the external shuffle service becomes
unusable.
I don't think of losing all the shuffle data to be as aggressive as
blacklisting the node, but maybe you're right, its pretty similar.
I guess I'm OK with this, though I'm still not sure when I'd tell users to
turn this on.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
