[SPARK-32794][SS] Fixed rare corner case error in micro-batch engine with some stateful queries + no-data-batches + V1 sources #29700
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
…with some stateful queries + no-data-batches + V1 sources
Make MicroBatchExecution explicitly call `getBatch` when the start and end offsets are the same.
Structured Streaming micro-batch engine has the contract with V1 data sources that, after a restart, it will call `source.getBatch()` on the last batch attempted before the restart. However, a very rare combination of sequences violates this contract. It occurs only when
- The streaming query has specific types of stateful operations with watermarks (e.g., aggregation in append, mapGroupsWithState with timeouts).
- These queries can execute a batch even without new data when the previous updates the watermark and the stateful ops are such that the new watermark can cause new output/cleanup. Such batches are called no-data-batches.
- The last batch before termination was an incomplete no-data-batch. Upon restart, the micro-batch engine fails to call `source.getBatch` when attempting to re-execute the incomplete no-data-batch.
This occurs because no-data-batches has the same and end offsets, and when a batch is executed, if the start and end offset is same then calling `source.getBatch` is skipped as it is assumed the generated plan will be empty. This only affects V1 data sources which rely on this invariant to detect in the source whether the query is being started from scratch or restarted.
No
New unit test with a mock v1 source that fails without the fix.
Closes apache#29696 from tdas/SPARK-32794-3.0.
Authored-by: Tathagata Das <[email protected]>
Signed-off-by: Tathagata Das <[email protected]>
|
Test build #128472 has finished for PR 29700 at commit
|
asfgit
pushed a commit
that referenced
this pull request
Sep 11, 2020
…with some stateful queries + no-data-batches + V1 sources
### What changes were proposed in this pull request?
Make MicroBatchExecution explicitly call `getBatch` when the start and end offsets are the same.
### Why are the changes needed?
Structured Streaming micro-batch engine has the contract with V1 data sources that, after a restart, it will call `source.getBatch()` on the last batch attempted before the restart. However, a very rare combination of sequences violates this contract. It occurs only when
- The streaming query has specific types of stateful operations with watermarks (e.g., aggregation in append, mapGroupsWithState with timeouts).
- These queries can execute a batch even without new data when the previous updates the watermark and the stateful ops are such that the new watermark can cause new output/cleanup. Such batches are called no-data-batches.
- The last batch before termination was an incomplete no-data-batch. Upon restart, the micro-batch engine fails to call `source.getBatch` when attempting to re-execute the incomplete no-data-batch.
This occurs because no-data-batches has the same and end offsets, and when a batch is executed, if the start and end offset is same then calling `source.getBatch` is skipped as it is assumed the generated plan will be empty. This only affects V1 data sources which rely on this invariant to detect in the source whether the query is being started from scratch or restarted.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
New unit test with a mock v1 source that fails without the fix.
Closes #29700 from tdas/SPARK-32794-2.4.
Authored-by: Tathagata Das <[email protected]>
Signed-off-by: Tathagata Das <[email protected]>
Member
|
Thank you, @tdas and @zsxwing . |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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?
Make MicroBatchExecution explicitly call
getBatchwhen the start and end offsets are the same.Why are the changes needed?
Structured Streaming micro-batch engine has the contract with V1 data sources that, after a restart, it will call
source.getBatch()on the last batch attempted before the restart. However, a very rare combination of sequences violates this contract. It occurs only whensource.getBatchwhen attempting to re-execute the incomplete no-data-batch.This occurs because no-data-batches has the same and end offsets, and when a batch is executed, if the start and end offset is same then calling
source.getBatchis skipped as it is assumed the generated plan will be empty. This only affects V1 data sources which rely on this invariant to detect in the source whether the query is being started from scratch or restarted.Does this PR introduce any user-facing change?
No
How was this patch tested?
New unit test with a mock v1 source that fails without the fix.