Skip to content

[SPARK-8029] robust shuffle writer (for 1.5 branch) - #9686

Closed
davies wants to merge 3 commits into
apache:branch-1.5from
davies:writer_1.5
Closed

[SPARK-8029] robust shuffle writer (for 1.5 branch)#9686
davies wants to merge 3 commits into
apache:branch-1.5from
davies:writer_1.5

Conversation

@davies

Copy link
Copy Markdown
Contributor

Currently, all the shuffle writer will write to target path directly, the file could be corrupted by other attempt of the same partition on the same executor. They should write to temporary file then rename to target path, as what we do in output committer. In order to make the rename atomic, the temporary file should be created in the same local directory (FileSystem).

This PR is based on #9214 , thanks to @squito

@daviesdavies changed the title [SPARK-8029] robust shuffle writer[SPARK-8029] robust shuffle writer (for 1.5 branch)Nov 13, 2015
@davies

Copy link
Copy Markdown
ContributorAuthor

cc @andrewor14

@SparkQA

Copy link
Copy Markdown

Test build #45841 timed out for PR 9686 at commit 7e83298 after a configured wait of 175m.

@SparkQA

Copy link
Copy Markdown

Test build #45847 has finished for PR 9686 at commit a6d569e.

  • This patch fails Spark unit tests.
  • This patch merges cleanly.
  • This patch adds no public classes.

@andrewor14

Copy link
Copy Markdown
Contributor

retest this please

Copy link
Copy Markdown
ContributorAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@aarondav Do you remember why we did not enable consolidation in this test? Since the title is "consolidated shuffle can write to shuffle group without messing existing offsets/lengths" (introduced in #1678)

@SparkQA

Copy link
Copy Markdown

Test build #45865 has finished for PR 9686 at commit a6d569e.

  • This patch fails Spark unit tests.
  • This patch merges cleanly.
  • This patch adds no public classes.

@andrewor14

Copy link
Copy Markdown
Contributor

retest this please

@SparkQA

Copy link
Copy Markdown

Test build #45873 has finished for PR 9686 at commit e95fcf5.

  • This patch passes all tests.
  • This patch merges cleanly.
  • This patch adds no public classes.

@andrewor14

Copy link
Copy Markdown
Contributor

merged into 1.5

asfgit pushed a commit that referenced this pull request Nov 13, 2015
Currently, all the shuffle writer will write to target path directly, the file could be corrupted by other attempt of the same partition on the same executor. They should write to temporary file then rename to target path, as what we do in output committer. In order to make the rename atomic, the temporary file should be created in the same local directory (FileSystem).
This PR is based on #9214 , thanks to squito
Author: Davies Liu <davies@databricks.com>
Closes#9686 from davies/writer_1.5 and squashes the following commits:
e95fcf5 [Davies Liu] fix test
a6d569e [Davies Liu] fix consolidate
7e83298 [Davies Liu] robust shuffle writer
@SparkQA

Copy link
Copy Markdown

Test build #45879 has finished for PR 9686 at commit e95fcf5.

  • This patch fails PySpark unit tests.
  • This patch merges cleanly.
  • This patch adds no public classes.

@andrewor14

Copy link
Copy Markdown
Contributor

@davies can you close this

@daviesdavies closed this Nov 13, 2015
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants

@davies@SparkQA@andrewor14