GitHub user ehnalis opened a pull request:
https://github.com/apache/spark/pull/8979
[SPARK-10797] RDD's coalesce should not write out the temporary key
I think we have the following options to solve this problem:
1. [General, but too complex] Create a new ShuffledRDD, add support to
shuffle system to read values instead of key-value pairs.
2. [General, too complex] Create a new CoalescedRDD, that bypasses
ShuffledRDD and reads from shuffle-layer directly - still need to add support
to read values.
3. [Minimal impact on code] Add option for shuffle-layer to write out
values only and cast the key-value iterator at read-site to Iterator[T].
I've implemented and I'm using the 3rd option. You will experience speed-up
based on the size of current payload (values).
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/ehnalis/spark coalesce
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/8979.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 #8979
----
commit 6fc5cf4bcef31c0c3d28b9ef011ca5de9f29c873
Author: ehnalis <[email protected]>
Date: 2015-10-05T13:14:34Z
Fixed tests.
commit 178a04ebaef9f7b0a03c87589caf58ead93baf30
Author: ehnalis <[email protected]>
Date: 2015-10-05T13:27:09Z
[SPARK-10797] RDD's coalesce should not write out the temporary key
commit 2b5f77b5333b93bb62f311bed37694e7f3ab0fd1
Author: ehnalis <[email protected]>
Date: 2015-10-05T13:29:42Z
[SPARK-10797] RDD's coalesce should not write out the temporary key
----
---
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]
