Github user kiszk commented on the issue: https://github.com/apache/spark/pull/11956 I saw PR #13899. I understood there are two design points for now. Hopefully, no more points :) 1. `CachedBatch` must be serializable. 2. According to my experiments, we should make a class for a `CachedBatch` simple. For 1, current ColumnarBatch is not serializable. For 2, at [this line](https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/storage/memory/MemoryStore.scala#L216), `.next()` returns `CachedBatch`, and then `+=` involves call [`SizeEstimator.estimate()`](https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/util/SizeEstimator.scala#L69). If this method takes longer time for a complicated `CacheBatch` class, the performance is degraded. One of example is `auto_join??` in [this test](https://github.com/apache/spark/blob/master/sql/hive/compatibility/src/test/scala/org/apache/spark/sql/hive/execution/HiveCompatibilitySuite.scala).
--- 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