Skip to content

Remove CoalescePartitions insertion from Joins #15570

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 19 commits into from
Apr 7, 2025

Conversation

ctsk
Copy link
Contributor

@ctsk ctsk commented Apr 3, 2025

Continuing #15476 and #15418

Rationale for this change + What changes are included in this PR?

The logic of HashJoin's execute implements differs from the other operators - The recursive call to the execute step of the build side are delayed until collect_left_join is poll'ed. This PR changes that to make it more alike to the standard implementations of execute.

Are these changes tested?

No changes in behaviour are expected.

Are there any user-facing changes?

No

Notable changes:

  • Fixed the bug that caused test failures (partition 0 of the left side was executed multiple times)
  • Applied the same changes the nested_loop_join and cross_join

Copy link
Contributor

@berkaysynnada berkaysynnada left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The changes are LGTM, thank you @ctsk for tracking and finalizing this issue

@alamb
Copy link
Contributor

alamb commented Apr 6, 2025

I wonder if there is any way to write some tests for this (perhaps via EXPLAIN in .slt tests to demonstrate that the unecessary exec is removed)

@alamb
Copy link
Contributor

alamb commented Apr 6, 2025

🤖 ./gh_compare_branch.sh Benchmark Script Running
Linux aal-dev 6.8.0-1016-gcp #18-Ubuntu SMP Fri Oct 4 22:16:29 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux
Comparing remove-hj-coalesce (995a95c) to 0b061be diff
Benchmarks: tpch_mem clickbench_partitioned clickbench_extended
Results will be posted here when complete

@berkaysynnada
Copy link
Contributor

berkaysynnada commented Apr 6, 2025

I wonder if there is any way to write some tests for this (perhaps via EXPLAIN in .slt tests to demonstrate that the unecessary exec is removed)

I don't think a redundant exec can be removed since if the input is 1 partition, it was already not introducing a new one, and what we see is input is always 1 (enforce distribution does its job 👏🏻), because we don't get any plan changes

@alamb
Copy link
Contributor

alamb commented Apr 6, 2025

I see -- this is code simplification 👍

@alamb
Copy link
Contributor

alamb commented Apr 6, 2025

🤖: Benchmark completed

Details

Comparing HEAD and remove-hj-coalesce
--------------------
Benchmark clickbench_extended.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Query        ┃       HEAD ┃ remove-hj-coalesce ┃    Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ QQuery 0     │  1983.64ms │          1974.40ms │ no change │
│ QQuery 1     │   699.38ms │           708.13ms │ no change │
│ QQuery 2     │  1421.28ms │          1456.67ms │ no change │
│ QQuery 3     │   708.08ms │           717.37ms │ no change │
│ QQuery 4     │  1510.24ms │          1496.34ms │ no change │
│ QQuery 5     │ 17349.43ms │         17050.26ms │ no change │
│ QQuery 6     │  6706.21ms │          6713.50ms │ no change │
└──────────────┴────────────┴────────────────────┴───────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                 ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                 │ 30378.27ms │
│ Total Time (remove-hj-coalesce)   │ 30116.68ms │
│ Average Time (HEAD)               │  4339.75ms │
│ Average Time (remove-hj-coalesce) │  4302.38ms │
│ Queries Faster                    │          0 │
│ Queries Slower                    │          0 │
│ Queries with No Change            │          7 │
└───────────────────────────────────┴────────────┘
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃       HEAD ┃ remove-hj-coalesce ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0     │     1.99ms │             1.92ms │     no change │
│ QQuery 1     │    35.20ms │            34.74ms │     no change │
│ QQuery 2     │    91.74ms │            91.34ms │     no change │
│ QQuery 3     │   102.29ms │           101.18ms │     no change │
│ QQuery 4     │   744.08ms │           740.73ms │     no change │
│ QQuery 5     │   877.28ms │           865.60ms │     no change │
│ QQuery 6     │     1.91ms │             1.94ms │     no change │
│ QQuery 7     │    40.83ms │            39.88ms │     no change │
│ QQuery 8     │   936.92ms │           953.98ms │     no change │
│ QQuery 9     │  1256.02ms │          1281.07ms │     no change │
│ QQuery 10    │   275.47ms │           273.32ms │     no change │
│ QQuery 11    │   301.56ms │           305.70ms │     no change │
│ QQuery 12    │   937.22ms │           942.74ms │     no change │
│ QQuery 13    │  1273.43ms │          1275.80ms │     no change │
│ QQuery 14    │   881.06ms │           881.44ms │     no change │
│ QQuery 15    │  1070.13ms │          1068.40ms │     no change │
│ QQuery 16    │  1803.13ms │          1794.14ms │     no change │
│ QQuery 17    │  1638.06ms │          1651.44ms │     no change │
│ QQuery 18    │  3170.64ms │          3178.40ms │     no change │
│ QQuery 19    │    85.90ms │            87.13ms │     no change │
│ QQuery 20    │  1144.96ms │          1141.99ms │     no change │
│ QQuery 21    │  1380.02ms │          1353.25ms │     no change │
│ QQuery 22    │  2504.46ms │          2484.34ms │     no change │
│ QQuery 23    │  8618.09ms │          8678.11ms │     no change │
│ QQuery 24    │   483.76ms │           477.05ms │     no change │
│ QQuery 25    │   393.25ms │           397.38ms │     no change │
│ QQuery 26    │   550.26ms │           543.41ms │     no change │
│ QQuery 27    │  1738.81ms │          1705.99ms │     no change │
│ QQuery 28    │ 12715.67ms │         12666.45ms │     no change │
│ QQuery 29    │   530.60ms │           550.06ms │     no change │
│ QQuery 30    │   868.66ms │           848.77ms │     no change │
│ QQuery 31    │   913.96ms │           899.26ms │     no change │
│ QQuery 32    │  2762.98ms │          2813.21ms │     no change │
│ QQuery 33    │  3447.68ms │          3417.70ms │     no change │
│ QQuery 34    │  3430.75ms │          3448.18ms │     no change │
│ QQuery 35    │  1331.45ms │          1316.45ms │     no change │
│ QQuery 36    │   159.48ms │           154.87ms │     no change │
│ QQuery 37    │    88.14ms │            85.47ms │     no change │
│ QQuery 38    │   149.12ms │           153.57ms │     no change │
│ QQuery 39    │   227.91ms │           228.93ms │     no change │
│ QQuery 40    │    77.19ms │            79.93ms │     no change │
│ QQuery 41    │    76.35ms │            72.23ms │ +1.06x faster │
│ QQuery 42    │    65.96ms │            67.20ms │     no change │
└──────────────┴────────────┴────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                 ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                 │ 59184.35ms │
│ Total Time (remove-hj-coalesce)   │ 59154.69ms │
│ Average Time (HEAD)               │  1376.38ms │
│ Average Time (remove-hj-coalesce) │  1375.69ms │
│ Queries Faster                    │          1 │
│ Queries Slower                    │          0 │
│ Queries with No Change            │         42 │
└───────────────────────────────────┴────────────┘
--------------------
Benchmark tpch_mem_sf1.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃     HEAD ┃ remove-hj-coalesce ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1     │ 121.75ms │           123.41ms │     no change │
│ QQuery 2     │  24.15ms │            24.22ms │     no change │
│ QQuery 3     │  35.77ms │            36.30ms │     no change │
│ QQuery 4     │  21.25ms │            21.42ms │     no change │
│ QQuery 5     │  57.69ms │            58.52ms │     no change │
│ QQuery 6     │   8.21ms │             8.27ms │     no change │
│ QQuery 7     │ 110.68ms │           104.15ms │ +1.06x faster │
│ QQuery 8     │  26.96ms │            26.31ms │     no change │
│ QQuery 9     │  63.25ms │            63.22ms │     no change │
│ QQuery 10    │  61.03ms │            61.83ms │     no change │
│ QQuery 11    │  13.05ms │            13.21ms │     no change │
│ QQuery 12    │  37.51ms │            39.51ms │  1.05x slower │
│ QQuery 13    │  29.16ms │            29.59ms │     no change │
│ QQuery 14    │   9.79ms │            10.05ms │     no change │
│ QQuery 15    │  25.66ms │            25.54ms │     no change │
│ QQuery 16    │  24.00ms │            24.33ms │     no change │
│ QQuery 17    │  98.05ms │            96.97ms │     no change │
│ QQuery 18    │ 249.89ms │           249.44ms │     no change │
│ QQuery 19    │  28.19ms │            28.38ms │     no change │
│ QQuery 20    │  39.71ms │            40.64ms │     no change │
│ QQuery 21    │ 176.95ms │           169.94ms │     no change │
│ QQuery 22    │  17.44ms │            18.33ms │  1.05x slower │
└──────────────┴──────────┴────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                 ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                 │ 1280.14ms │
│ Total Time (remove-hj-coalesce)   │ 1273.58ms │
│ Average Time (HEAD)               │   58.19ms │
│ Average Time (remove-hj-coalesce) │   57.89ms │
│ Queries Faster                    │         1 │
│ Queries Slower                    │         2 │
│ Queries with No Change            │        19 │
└───────────────────────────────────┴───────────┘

@alamb
Copy link
Contributor

alamb commented Apr 6, 2025

TLDR is benchmark results look good to me -- thanks @ctsk !

Copy link
Contributor

@alamb alamb left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you @ctsk and @berkaysynnada

@@ -282,6 +282,13 @@ impl ExecutionPlan for CrossJoinExec {
partition: usize,
context: Arc<TaskContext>,
) -> Result<SendableRecordBatchStream> {
if self.left.output_partitioning().partition_count() != 1 {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

love it

@alamb alamb merged commit 3fe300c into apache:main Apr 7, 2025
27 checks passed
nirnayroy pushed a commit to nirnayroy/datafusion that referenced this pull request May 2, 2025
* refactor: Catch PartitionMode:Auto in execute explicitly

* refactor(hash_join): Move coalesce logic into function

* refactor(hash_join): Move coalesce logic out of collect_left_input

* refactor(hash_join): Execute build side earlier

* chore(hash_join): Drop unnecessary clippy hint \o/

* Back out "refactor: Catch PartitionMode:Auto in execute explicitly"

This backs out commit 4346cb8.

* Remove CoalescePartitions from CollectLeft-HashJoins

* Fix imports

* Fix tests

* Check CollectLeft-HJ distribution in execute

* Fix repeated execute; propagate execute error

* Remove CoalescePartitions from nested_loop_join

* Execute left side earlier in nested_loop_join

* Remove CoalescePartitions from cross_join

* Execute left side earlier in cross_join

* Remove unused import

* Fix docs

* Remove dead comment

* Throw error on unexpected distribution
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants