[SPARK-25331][SS] Make FileStreamSink ignore partitions of batches that have already been written to file system #22331
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Outdated suggestions cannot be applied.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?
Reproduce File Sink duplication in driver failure scenario to help understanding the situation.
Propose a new StagingFileCommitProtocol that creates the target files in a
staging
subdirectory and upon job commit it moves all the files to the target directory. The target file names of a task will be the same in 2 different runs. This way a potential source for duplication (when same content is placed into 2 files with different names) is eliminated.How was this patch tested?
Created specific unit test for the new protocol: StagingFileCommitProtocolSuite
Ran the test that reproduced the problem: FileStreamSinkUnitSuite
Made FileStreamStressSuite more strict to demand exactly once delivery.
Tested on a 4 machine cluster sending 30000 messages 20 times killing the driver. Each message was delivered exactly once.
Ran tests for sql with sbt.