You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@arrow.apache.org by "mingmwang (via GitHub)" <gi...@apache.org> on 2023/04/12 09:38:11 UTC
[GitHub] [arrow-datafusion] mingmwang opened a new pull request, #5973: Row accumulator support Decimal type
mingmwang opened a new pull request, #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973
# Which issue does this PR close?
<!--
We generally require a GitHub issue to be filed for all bug fixes and enhancements and this helps us generate change logs for our releases. You can link an issue to this PR using the GitHub syntax. For example `Closes #123` indicates that this PR will close issue #123.
-->
Closes #5892.
# Rationale for this change
Improve the Aggregate performance for decimal type
<!--
Why are you proposing this change? If this is already explained clearly in the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand your changes and offer better suggestions for fixes.
-->
# What changes are included in this PR?
<!--
There is no need to duplicate the description in the issue here but it is sometimes worth providing a summary of the individual changes in this PR.
-->
# Are these changes tested?
I had test this on my local Mac, for TPCH-q17, there is at about 30% improvement.
After this PR:
Running benchmarks with the following options: DataFusionBenchmarkOpt { query: Some(17), debug: false, iterations: 3, partitions: 1, batch_size: 8192, path: "./parquet_data", file_format: "parquet", mem_table: false, output_path: None, disable_statistics: true, enable_scheduler: false }
Query 17 iteration 0 took 1978.5 ms and returned 1 rows
Query 17 iteration 1 took 1941.8 ms and returned 1 rows
Query 17 iteration 2 took 1952.6 ms and returned 1 rows
Query 17 avg time: 1957.62 ms
<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code
If tests are not included in your PR, please explain why (for example, are they covered by existing tests)?
-->
# Are there any user-facing changes?
<!--
If there are user-facing changes then we may require documentation to be updated before approving the PR.
-->
<!--
If there are any breaking changes to public APIs, please add the `api change` label.
-->
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] mingmwang commented on pull request #5973: Row accumulator support Decimal type
Posted by "mingmwang (via GitHub)" <gi...@apache.org>.
mingmwang commented on PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973#issuecomment-1504972362
@andygrove @Dandandan @yahoNanJing
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] mingmwang commented on a diff in pull request #5973: Row `AVG` accumulator support Decimal type
Posted by "mingmwang (via GitHub)" <gi...@apache.org>.
mingmwang commented on code in PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973#discussion_r1164775086
##########
datafusion/physical-expr/src/aggregate/average.rs:
##########
@@ -298,16 +305,40 @@ impl RowAccumulator for AvgRowAccumulator {
}
fn evaluate(&self, accessor: &RowAccessor) -> Result<ScalarValue> {
- assert_eq!(self.sum_datatype, DataType::Float64);
- Ok(match accessor.get_u64_opt(self.state_index()) {
- None => ScalarValue::Float64(None),
- Some(0) => ScalarValue::Float64(None),
- Some(n) => ScalarValue::Float64(
- accessor
- .get_f64_opt(self.state_index() + 1)
- .map(|f| f / n as f64),
- ),
- })
+ match self.sum_datatype {
+ DataType::Decimal128(p, s) => {
+ match accessor.get_u64_opt(self.state_index()) {
+ None => Ok(ScalarValue::Decimal128(None, p, s)),
+ Some(0) => Ok(ScalarValue::Decimal128(None, p, s)),
Review Comment:
I will do some test on this today.
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] mingmwang commented on pull request #5973: Row `AVG` accumulator support Decimal type
Posted by "mingmwang (via GitHub)" <gi...@apache.org>.
mingmwang commented on PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973#issuecomment-1506431739
For query 18, I think the plan is problematic, it is the Join order and build side selection, the bottleneck is not the Aggregations.
```
=== Physical plan with metrics ===
SortExec: expr=[o_totalprice@4 DESC,o_orderdate@3 ASC NULLS LAST], metrics=[output_rows=57, elapsed_compute=11.461µs, spill_count=0, spilled_bytes=0]
AggregateExec: mode=Single, gby=[c_name@1 as c_name, c_custkey@0 as c_custkey, o_orderkey@2 as o_orderkey, o_orderdate@4 as o_orderdate, o_totalprice@3 as o_totalprice], aggr=[SUM(lineitem.l_quantity)], metrics=[output_rows=57, elapsed_compute=49.165µs, spill_count=0, spilled_bytes=0, mem_used=0]
CoalesceBatchesExec: target_batch_size=8192, metrics=[output_rows=399, elapsed_compute=2.681µs, spill_count=0, spilled_bytes=0, mem_used=0]
HashJoinExec: mode=CollectLeft, join_type=LeftSemi, on=[(Column { name: "o_orderkey", index: 2 }, Column { name: "l_orderkey", index: 0 })], metrics=[output_rows=456, input_rows=456, input_batches=2, build_input_batches=733, build_input_rows=6001215, output_batches=2, build_mem_used=806510152, build_time=676.460128ms, join_time=2.88546ms]
ProjectionExec: expr=[c_custkey@0 as c_custkey, c_name@1 as c_name, o_orderkey@2 as o_orderkey, o_totalprice@3 as o_totalprice, o_orderdate@4 as o_orderdate, l_quantity@6 as l_quantity], metrics=[output_rows=6001215, elapsed_compute=116.294µs, spill_count=0, spilled_bytes=0, mem_used=0]
CoalesceBatchesExec: target_batch_size=8192, metrics=[output_rows=6001215, elapsed_compute=50.883µs, spill_count=0, spilled_bytes=0, mem_used=0]
HashJoinExec: mode=CollectLeft, join_type=Inner, on=[(Column { name: "o_orderkey", index: 2 }, Column { name: "l_orderkey", index: 0 })], metrics=[output_rows=6001215, input_rows=6001215, input_batches=733, build_input_batches=184, build_input_rows=1500000, output_batches=733, build_mem_used=177373616, build_time=154.423989ms, join_time=268.907576ms]
ProjectionExec: expr=[c_custkey@0 as c_custkey, c_name@1 as c_name, o_orderkey@2 as o_orderkey, o_totalprice@4 as o_totalprice, o_orderdate@5 as o_orderdate], metrics=[output_rows=1500000, elapsed_compute=34.671µs, spill_count=0, spilled_bytes=0, mem_used=0]
CoalesceBatchesExec: target_batch_size=8192, metrics=[output_rows=1500000, elapsed_compute=14.97µs, spill_count=0, spilled_bytes=0, mem_used=0]
HashJoinExec: mode=CollectLeft, join_type=Inner, on=[(Column { name: "c_custkey", index: 0 }, Column { name: "o_custkey", index: 1 })], metrics=[output_rows=1500000, input_rows=1500000, input_batches=184, build_input_batches=19, build_input_rows=150000, output_batches=184, build_mem_used=14652720, build_time=7.173373ms, join_time=64.614167ms]
ParquetExec: limit=None, partitions={1 group: [[Users/mingmwang/gitrepo/apache/arrow-datafusion/benchmarks/parquet_data/customer/part-0.parquet]]}, projection=[c_custkey, c_name], metrics=[output_rows=150000, elapsed_compute=1ns, spill_count=0, spilled_bytes=0, mem_used=0, pushdown_rows_filtered=0, num_predicate_creation_errors=0, predicate_evaluation_errors=0, row_groups_pruned=0, bytes_scanned=566600, page_index_rows_filtered=0, pushdown_eval_time=2ns, time_elapsed_scanning_total=5.384791ms, time_elapsed_processing=5.024293ms, time_elapsed_scanning_until_data=2.83875ms, time_elapsed_opening=787.167µs, page_index_eval_time=2ns]
ParquetExec: limit=None, partitions={1 group: [[Users/mingmwang/gitrepo/apache/arrow-datafusion/benchmarks/parquet_data/orders/part-0.parquet]]}, projection=[o_orderkey, o_custkey, o_totalprice, o_orderdate], metrics=[output_rows=1500000, elapsed_compute=1ns, spill_count=0, spilled_bytes=0, mem_used=0, pushdown_rows_filtered=0, num_predicate_creation_errors=0, predicate_evaluation_errors=0, row_groups_pruned=0, bytes_scanned=13916402, page_index_rows_filtered=0, pushdown_eval_time=2ns, time_elapsed_scanning_total=112.261704ms, time_elapsed_processing=45.363041ms, time_elapsed_scanning_until_data=6.063625ms, time_elapsed_opening=549.5µs, page_index_eval_time=2ns]
ParquetExec: limit=None, partitions={1 group: [[Users/mingmwang/gitrepo/apache/arrow-datafusion/benchmarks/parquet_data/lineitem/part-0.parquet]]}, projection=[l_orderkey, l_quantity], metrics=[output_rows=6001215, elapsed_compute=1ns, spill_count=0, spilled_bytes=0, mem_used=0, pushdown_rows_filtered=0, num_predicate_creation_errors=0, predicate_evaluation_errors=0, row_groups_pruned=0, bytes_scanned=12170874, page_index_rows_filtered=0, pushdown_eval_time=2ns, time_elapsed_scanning_total=324.056333ms, time_elapsed_processing=50.1033ms, time_elapsed_scanning_until_data=2.464666ms, time_elapsed_opening=2.0565ms, page_index_eval_time=2ns]
ProjectionExec: expr=[l_orderkey@0 as l_orderkey], metrics=[output_rows=57, elapsed_compute=458ns, spill_count=0, spilled_bytes=0, mem_used=0]
CoalesceBatchesExec: target_batch_size=8192, metrics=[output_rows=57, elapsed_compute=19.342µs, spill_count=0, spilled_bytes=0, mem_used=0]
FilterExec: SUM(lineitem.l_quantity)@1 > Some(30000),25,2, metrics=[output_rows=57, elapsed_compute=1.093341ms, spill_count=0, spilled_bytes=0, mem_used=0]
AggregateExec: mode=Single, gby=[l_orderkey@0 as l_orderkey], aggr=[SUM(lineitem.l_quantity)], metrics=[output_rows=1500000, elapsed_compute=434.341183ms, spill_count=0, spilled_bytes=0, mem_used=0]
ParquetExec: limit=None, partitions={1 group: [[Users/mingmwang/gitrepo/apache/arrow-datafusion/benchmarks/parquet_data/lineitem/part-0.parquet]]}, projection=[l_orderkey, l_quantity], metrics=[output_rows=6001215, elapsed_compute=1ns, spill_count=0, spilled_bytes=0, mem_used=0, pushdown_rows_filtered=0, num_predicate_creation_errors=0, predicate_evaluation_errors=0, row_groups_pruned=0, bytes_scanned=12170874, page_index_rows_filtered=0, pushdown_eval_time=2ns, time_elapsed_scanning_total=443.876134ms, time_elapsed_processing=49.172503ms, time_elapsed_scanning_until_data=2.226959ms, time_elapsed_opening=844.959µs, page_index_eval_time=2ns]
```
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] mingmwang merged pull request #5973: Row `AVG` accumulator support Decimal type
Posted by "mingmwang (via GitHub)" <gi...@apache.org>.
mingmwang merged PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] alamb commented on a diff in pull request #5973: Row `AVG` accumulator support Decimal type
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb commented on code in PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973#discussion_r1164540227
##########
datafusion/physical-expr/src/aggregate/average.rs:
##########
@@ -298,16 +305,40 @@ impl RowAccumulator for AvgRowAccumulator {
}
fn evaluate(&self, accessor: &RowAccessor) -> Result<ScalarValue> {
- assert_eq!(self.sum_datatype, DataType::Float64);
- Ok(match accessor.get_u64_opt(self.state_index()) {
- None => ScalarValue::Float64(None),
- Some(0) => ScalarValue::Float64(None),
- Some(n) => ScalarValue::Float64(
- accessor
- .get_f64_opt(self.state_index() + 1)
- .map(|f| f / n as f64),
- ),
- })
+ match self.sum_datatype {
+ DataType::Decimal128(p, s) => {
+ match accessor.get_u64_opt(self.state_index()) {
+ None => Ok(ScalarValue::Decimal128(None, p, s)),
+ Some(0) => Ok(ScalarValue::Decimal128(None, p, s)),
Review Comment:
Why do we translate `0` --> `null` here? (it is also done for Float64 below)?
I see you are just following the existing pattern, but it seems like this could be incorrect?
Maybe we could add a test that calls `AVG` on (`-1` and `1`) to see if we get `0` or `NULL`
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] yahoNanJing commented on a diff in pull request #5973: Row `AVG` accumulator support Decimal type
Posted by "yahoNanJing (via GitHub)" <gi...@apache.org>.
yahoNanJing commented on code in PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973#discussion_r1164923140
##########
datafusion/physical-expr/src/aggregate/average.rs:
##########
@@ -298,16 +305,40 @@ impl RowAccumulator for AvgRowAccumulator {
}
fn evaluate(&self, accessor: &RowAccessor) -> Result<ScalarValue> {
- assert_eq!(self.sum_datatype, DataType::Float64);
- Ok(match accessor.get_u64_opt(self.state_index()) {
- None => ScalarValue::Float64(None),
- Some(0) => ScalarValue::Float64(None),
- Some(n) => ScalarValue::Float64(
- accessor
- .get_f64_opt(self.state_index() + 1)
- .map(|f| f / n as f64),
- ),
- })
+ match self.sum_datatype {
+ DataType::Decimal128(p, s) => {
+ match accessor.get_u64_opt(self.state_index()) {
+ None => Ok(ScalarValue::Decimal128(None, p, s)),
+ Some(0) => Ok(ScalarValue::Decimal128(None, p, s)),
Review Comment:
Hi @alamb, the value at the `state_index` is for the count rather than the sum. When the count is 0, for the average, it should be NULL.
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] alamb commented on pull request #5973: Row `AVG` accumulator support Decimal type
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb commented on PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973#issuecomment-1507406719
🎉
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [arrow-datafusion] mingmwang commented on pull request #5973: Row `AVG` accumulator support Decimal type
Posted by "mingmwang (via GitHub)" <gi...@apache.org>.
mingmwang commented on PR #5973:
URL: https://github.com/apache/arrow-datafusion/pull/5973#issuecomment-1507120900
> Thanks @mingmwang @yahoNanJing! Looks great to me.
>
> Could we also add two roundtrip (`DecimalArray` -> `vec<u8>` -> `DecimalArray`) tests in `datafusion/row/src/lib.rs` for the newly introduced Decimal type? One for a null-free case and one for a nullable case.
Sure, I will add the two tests in a following PR.
--
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.
To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org