Github user viirya commented on the pull request:
https://github.com/apache/spark/pull/5475#issuecomment-92664701
i just found that we can only allocate the accumulator for first one
`InMemoryRelation` instead of for all following `InMemoryRelation`. Because the
following relations only need to use the pre-computed statistics of the first
relation.
The accumulators in `InMemoryColumnarTableScan` are harder to completely
removed if we want to collect these statistics, because these statistics are
computed during the RDD operations. Accumulator seems the only method we can
pass the information out. Currently I still apply a configuration to control if
we want to allocate accumulators. But now I put these accumulators in an array
in `InMemoryRelation`, instead of a global map before. Of course this array
needs to be propagated to following relations because
`InMemoryColumnarTableScan` is depending on later relations not the original
one.
---
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]
