tdas opened a new pull request #27373: [SPARK-30657] [SPARK-30658] [SS] Fixed 
two bugs in streaming limits
URL: https://github.com/apache/spark/pull/27373
 
 
   <!--
   Thanks for sending a pull request!  Here are some tips for you:
     1. If this is your first time, please read our contributor guidelines: 
https://spark.apache.org/contributing.html
     2. Ensure you have added or run the appropriate tests for your PR: 
https://spark.apache.org/developer-tools.html
     3. If the PR is unfinished, add '[WIP]' in your PR title, e.g., 
'[WIP][SPARK-XXXX] Your PR title ...'.
     4. Be sure to keep the PR description updated to reflect all changes.
     5. Please write your PR title to summarize what this PR proposes.
     6. If possible, provide a concise example to reproduce the issue for a 
faster review.
   -->
   
   ### What changes were proposed in this pull request?
   This PR solves two bugs related to streaming limits
   
   **Bug 1 (SPARK-30658)**: Limit before a streaming aggregate (i.e. 
`df.limit(5).groupBy().count()`) in complete mode was not being planned as a 
streaming limit. The planner rule planned a logical limit with a stateful 
streaming limit plan only if the query is in append mode. As a result, instead 
of allowing max 5 rows across batches, the planned streaming query was allowing 
5 rows in every batch thus producing incorrect results.
   
   **Solution**: Change the planner rule to plan the logical limit with a 
streaming limit plan even when the query is in complete mode if the logical 
limit has no stateful operator before it.
   
   **Bug 2 (SPARK-30657)**: `LocalLimitExec` does not consume the iterator of 
the child plan. So if there is a limit after a stateful operator like streaming 
dedup in append mode (e.g. `df.dropDuplicates().limit(5)`), the state changes 
of streaming duplicate may not be committed (most stateful ops commit state 
changes only after the generated iterator is fully consumed).
   
   **Solution**: Change the planner rule to always use a new 
`StreamingLocalLimitExec` which always fully consumes the iterator. This is the 
safest thing to do. However, this will introduce a performance regression as 
consuming the iterator is extra work. To minimize this performance impact, add 
an additional post-planner optimization rule to replace 
`StreamingLocalLimitExec` with `LocalLimitExec` when there is no stateful 
operator before the limit that could be affected by it.
   
   ### Does this PR introduce any user-facing change?
   No
   
   ### How was this patch tested?
   Updated incorrect unit tests and added new ones

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to