GitHub user peter-toth opened a pull request:
https://github.com/apache/spark/pull/22603
SPARK-25062: clean up BlockLocations in InMemoryFileIndex
## What changes were proposed in this pull request?
`InMemoryFileIndex` caches `FileStatus` objects to paths. Each `FileStatus`
object can contain several `BlockLocations`. Depending on the parallel
discovery threshold (`spark.sql.sources.parallelPartitionDiscovery.threshold`)
the file listing can happen on the driver or on the executors. If the listing
happens on the executors the block location objects are converted to simple
`BlockLocation` objects to ensure serialization requirements. If it happens on
the driver then there is no conversion and depending on the file system a
`BlockLocation` object can be a subclass like `HdfsBlockLocation` and consume
more memory. This PR adds the conversion to the latter case and decreases
memory consumption.
## How was this patch tested?
Added unit test.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/peter-toth/spark SPARK-25062
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/22603.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #22603
----
commit 5735dba9a43c336843ccc531831105d6c23b4586
Author: Peter Toth <peter.toth@...>
Date: 2018-09-30T13:01:36Z
SPARK-25062: clean up BlockLocations in InMemoryFileIndex
----
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
