[SPARK-12717][PYTHON][BRANCH-2.1] Adding thread-safe broadcast pickle registry #18825
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added regression test for multithreaded broadcast pickle
Member
Author
|
backport for 2.1 |
Member
|
LGTM if it passes the tests. |
BryanCutler
changed the title
Member
Author
|
Since it seemed like some users in the JIRA might be using 2.1, I made this PR too. I think this is pretty low risk also, and it went in without any conflicts. But up to you if you're ok with this one also, as long as tests pass. |
|
Test build #80181 has finished for PR 18825 at commit
|
asfgit
pushed a commit
that referenced
this pull request
Aug 3, 2017
… registry ## What changes were proposed in this pull request? When using PySpark broadcast variables in a multi-threaded environment, `SparkContext._pickled_broadcast_vars` becomes a shared resource. A race condition can occur when broadcast variables that are pickled from one thread get added to the shared ` _pickled_broadcast_vars` and become part of the python command from another thread. This PR introduces a thread-safe pickled registry using thread local storage so that when python command is pickled (causing the broadcast variable to be pickled and added to the registry) each thread will have their own view of the pickle registry to retrieve and clear the broadcast variables used. ## How was this patch tested? Added a unit test that causes this race condition using another thread. Author: Bryan Cutler <[email protected]> Closes #18825 from BryanCutler/pyspark-bcast-threadsafe-SPARK-12717-2_1.
Member
|
Merged into branch-2.1. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What changes were proposed in this pull request?
When using PySpark broadcast variables in a multi-threaded environment,
SparkContext._pickled_broadcast_varsbecomes a shared resource. A race condition can occur when broadcast variables that are pickled from one thread get added to the shared_pickled_broadcast_varsand become part of the python command from another thread. This PR introduces a thread-safe pickled registry using thread local storage so that when python command is pickled (causing the broadcast variable to be pickled and added to the registry) each thread will have their own view of the pickle registry to retrieve and clear the broadcast variables used.How was this patch tested?
Added a unit test that causes this race condition using another thread.