Issue #6272: Window Scaled Repartitioning #6366
Merged
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 2 commits
February 19, 2023 16:19
There is no need to copy the old partitioned data when repartitioning. But if we have started to combine, then repartitioning will duplicate data, so don't resize then.
Closed
2 tasks
|
Thanks for the PR! This still does not seem to solve the problem completely, though. If I add a loop to the test: # name: test/sql/window/test_window_repartition.test_slow
# description: Window reparitioning at scale
# group: [window]
statement ok
PRAGMA enable_verification
statement ok
create table df as
select d, i v1
from
range(date '2017-01-01', date '2020-12-31', interval '1' day) t(d),
range(3000) i
;
loop i 0 100
query I
select count(*)
from (
select percent_rank() over (partition by d order by v1) as rank_v1
from df
);
----
4380000
endloopThen running it I eventually get an off-by-a small amount still. This amount always seems to be less than the vector size, so perhaps something happening with a resize while a vector is already in-flight? |
|
Disabling all in-flight resizing seems to fix the issue, but this may be too heavy-handed of a solution: diff --git a/src/execution/operator/aggregate/physical_window.cpp b/src/execution/operator/aggregate/physical_window.cpp
index fbd7865f02..958c574efe 100644
--- a/src/execution/operator/aggregate/physical_window.cpp
+++ b/src/execution/operator/aggregate/physical_window.cpp
@@ -170,7 +170,7 @@ private:
void WindowGlobalSinkState::ResizeGroupingData(idx_t cardinality) {
// Have we started to combine? Then just live with it.
- if (grouping_data && !grouping_data->GetPartitions().empty()) {
+ if (grouping_data) {
return;
}
// Is the average partition size too large? |
|
I was able to reproduce this locally, and this seems to fix it: void PartitionedColumnData::FlushAppendState(PartitionedColumnDataAppendState &state) {
for (idx_t i = 0; i < state.partition_buffers.size(); i++) {
auto &partition_buffer = *state.partition_buffers[i];
if (partition_buffer.size() > 0) {
partitions[i]->Append(partition_buffer);
partition_buffer.Reset(); // Add this line!
}
}
} The buffer needed to be reset. Nowhere else in the codebase did we use the append state after flushing it, so this was not a problem before. |
Fix reentry problem in PartitionedColumnData::FlushAppendState. Make stress test more brutal and reliable.
…into window-pct-range
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
There is no need to copy the old partitioned data when repartitioning.
But if we have started to combine, then repartitioning will duplicate data,
so don't resize then.