Github user joshdevins commented on the pull request:
https://github.com/apache/spark/pull/4593#issuecomment-75540678
I have the same concern as @dbtsai in his comment. Most consumers of this
API will already be caching their dataset before the learning phase. Without
user care, this will introduce effectively double caching (in terms of data
size of cached RDDs) and will cause many jobs to fail after upgrading by
exceeding available heap for RDD cache. Furthermore, we are making assumptions
about how to cache -- in-memory only in this case. Should we parameterise this?
Perhaps that will help send the message in the API that there is caching also
done before learning. (FWIW, in-memory is definitely the right default choice
here.)
See email thread on dev for my specific encountering of this bug:
http://mail-archives.apache.org/mod_mbox/spark-dev/201502.mbox/%3CCAH5MZvMBjqOST-9Nr9k1z1rUODfSiczr_fV9kwqDFqAMNLC2Zw%40mail.gmail.com%3E
---
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]
