planner, executor: enable inline projection for merge join #15463

lance6716 · 2020-03-18T09:13:35Z

What problem does this PR solve?

Problem Summary:

part of #14428

What is changed and how it works?

What's Changed:

enable inline projection for merge join

How it Works:

markChildrenUsedColsis applied to joiner.

Related changes

Check List

Tests

Unit test

Side effects

Release note

benchmark based on BenchmarkMergeJoinExec

two data sources with columns of type {longlong, double, varstring(5*2014)}

case 1

no modify:

BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4         	       1	2407966792 ns/op	58474936 B/op	    4070 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:3000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4           	       2	 514012846 ns/op	58471576 B/op	    4053 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:30,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4             	       2	 709402071 ns/op	62614336 B/op	    4007 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:330000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4         	       1	2381562052 ns/op	58466440 B/op	    4064 allocs/op

case 2

childrenUsedCols (and output schema) changed to:

{true, false, false},
{false, false, false},

significant improvement due to less column copy

BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4         	      10	 104559070 ns/op	  717604 B/op	    3940 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:3000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4           	      67	  18877980 ns/op	  727133 B/op	    3928 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:30,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4             	      54	  23297968 ns/op	 4873998 B/op	    3884 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:330000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4         	      10	 104550756 ns/op	  722180 B/op	    3940 allocs/op

case 3

childrenUsedCols (and output schema) changed to:

{true, false, false},
{false, true, false},

significant improvement, slightly slower than case 2, which is reasonable because output schema has an extra double column

BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4         	      10	 107155546 ns/op	  759140 B/op	    3957 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:3000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4           	      54	  22822921 ns/op	  761917 B/op	    3943 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:30,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4             	      37	  28515056 ns/op	 4910528 B/op	    3899 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:330000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)-4         	      10	 109264842 ns/op	  757220 B/op	    3954 allocs/op

planner/core/task.go

planner/cascades/testdata/integration_suite_out.json

lance6716 · 2020-03-18T10:27:33Z

/run-all-tests

SunRunAway

Rest LGTM

executor/benchmark_test.go

executor/executor_required_rows_test.go

…proj-merge-join

SunRunAway

LGTM.
Benchmark data is appreciated.

lance6716 · 2020-03-29T15:15:44Z

/bench +tpch

SunRunAway · 2020-03-29T15:20:42Z

Hi, @lance6716
You can take a look at BenchmarkMergeJoinExec and try to adjust to fit inline projection situation. This link may help you.

sre-bot · 2020-03-29T16:45:37Z

Benchmark Report

Run Sysbench Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: c8fa259b2e26eebc274a6637f1ee5dcc13102177
+++ tidb: 256a54df5a0609a561f438114233fbbe308222fd
tikv: ea0fd4d7688d1d462a94cfe4f7ab4d7cb0b30ab5
pd: a63cf42fe9687c0e580777a01facc4004da69b63
================================================================================
oltp_update_index:
    * QPS: 4346.24 ± 0.40% (std=12.51) delta: 0.29% (p=0.294)
    * Latency p50: 29.45 ± 0.41% (std=0.09) delta: -0.29%
    * Latency p99: 51.94 ± 0.00% (std=0.00) delta: -3.60%
            
oltp_insert:
    * QPS: 6917.50 ± 0.49% (std=23.50) delta: 0.12% (p=0.737)
    * Latency p50: 18.50 ± 0.47% (std=0.06) delta: -0.12%
    * Latency p99: 32.68 ± 2.26% (std=0.49) delta: 1.36%
            
oltp_read_write:
    * QPS: 15004.44 ± 0.30% (std=26.48) delta: 0.55% (p=0.573)
    * Latency p50: 170.97 ± 0.29% (std=0.30) delta: -0.55%
    * Latency p99: 342.80 ± 5.95% (std=13.45) delta: 2.04%
            
oltp_point_select:
    * QPS: 41951.37 ± 0.83% (std=208.27) delta: 1.65% (p=0.005)
    * Latency p50: 3.05 ± 0.90% (std=0.02) delta: -1.69%
    * Latency p99: 10.84 ± 0.00% (std=0.00) delta: 0.00%

lance6716 · 2020-03-30T05:41:22Z

Hi, @lance6716
You can take a look at BenchmarkMergeJoinExec and try to adjust to fit inline projection situation. This link may help you.

Done

sre-bot · 2020-03-30T10:59:19Z

Benchmark Report

Run Sysbench Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: 514b1809286a9d5c71f6cf257a40c12750d1f3f7
+++ tidb: 256a54df5a0609a561f438114233fbbe308222fd
tikv: be05858e9b68a4ceb5312cf9816074e199b2e293
pd: a63cf42fe9687c0e580777a01facc4004da69b63
================================================================================
oltp_update_index:
    * QPS: 5991.08 ± 0.58% (std=22.41) delta: 0.72% (p=0.207)
    * Latency p50: 21.37 ± 0.59% (std=0.08) delta: -0.19%
    * Latency p99: 40.73 ± 0.90% (std=0.37) delta: 0.90%
            
oltp_insert:
    * QPS: 4851.73 ± 0.10% (std=3.25) delta: -0.15% (p=0.324)
    * Latency p50: 26.38 ± 0.09% (std=0.02) delta: 0.13%
    * Latency p99: 47.06 ± 2.73% (std=0.95) delta: 5.25%
            
oltp_read_write:
    * QPS: 14958.91 ± 0.89% (std=86.66) delta: 0.03% (p=0.937)
    * Latency p50: 171.02 ± 0.41% (std=0.44) delta: -0.29%
    * Latency p99: 309.15 ± 7.36% (std=13.88) delta: -2.65%
            
oltp_point_select:
    * QPS: 42126.29 ± 0.61% (std=165.79) delta: -0.33% (p=0.314)
    * Latency p50: 3.04 ± 0.74% (std=0.01) delta: 0.14%
    * Latency p99: 10.46 ± 0.00% (std=0.00) delta: -0.90%
            
oltp_update_non_index:
    * QPS: 4746.37 ± 0.20% (std=5.66) delta: 0.04% (p=0.749)
    * Latency p50: 26.96 ± 0.20% (std=0.03) delta: 0.04%
    * Latency p99: 42.24 ± 2.73% (std=0.85) delta: -1.37%

SunRunAway · 2020-04-02T00:48:10Z

Hi, @lance6716
You can take a look at BenchmarkMergeJoinExec and try to adjust to fit inline projection situation. This link may help you.

Done

Hi, @lance6716
Well done.
Case 3 could persist in BenchmarkMergeJoinExec as a sub-bench.

sre-bot · 2020-04-02T02:14:38Z

Benchmark Report

Run Sysbench Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: 959eca8f3a395169bb5d5938f54fddab4d0750f5
+++ tidb: 256a54df5a0609a561f438114233fbbe308222fd
tikv: 3b11b678830a06e67bca0e68bf2fbe1c336a060e
pd: 1a801edaddbbe73ab17256ac354301c1ead02cd8
================================================================================
oltp_update_index:
    * QPS: 4314.46 ± 0.32% (std=9.25) delta: -0.44% (p=0.329)
    * Latency p50: 29.62 ± 0.22% (std=0.05) delta: 0.30%
    * Latency p99: 54.37 ± 4.55% (std=1.77) delta: 1.56%
            
oltp_insert:
    * QPS: 7036.44 ± 0.93% (std=47.70) delta: 1.50% (p=0.032)
    * Latency p50: 18.19 ± 0.92% (std=0.12) delta: -1.49%
    * Latency p99: 31.66 ± 0.90% (std=0.29) delta: -1.81%
            
oltp_read_write:
    * QPS: 14317.86 ± 0.77% (std=93.34) delta: 0.06% (p=0.909)
    * Latency p50: 179.19 ± 0.81% (std=1.15) delta: -0.04%
    * Latency p99: 335.39 ± 8.29% (std=18.07) delta: 4.11%
            
oltp_point_select:
    * QPS: 42667.50 ± 0.76% (std=194.90) delta: 2.12% (p=0.005)
    * Latency p50: 3.00 ± 0.67% (std=0.01) delta: -2.04%
    * Latency p99: 10.46 ± 0.00% (std=0.00) delta: -2.94%
            
oltp_update_non_index:
    * QPS: 4736.98 ± 0.35% (std=12.64) delta: -0.20% (p=0.315)
    * Latency p50: 27.02 ± 0.37% (std=0.07) delta: 0.19%
    * Latency p99: 42.24 ± 2.73% (std=0.85) delta: 1.84%

lance6716 · 2020-04-02T04:02:35Z

Hi, @lance6716
You can take a look at BenchmarkMergeJoinExec and try to adjust to fit inline projection situation. This link may help you.

Done

Hi, @lance6716
Well done.
Case 3 could persist in BenchmarkMergeJoinExec as a sub-bench.

goos: darwin
goarch: amd64
pkg: github.com/pingcap/tidb/executor
BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:false-4         	       1	2449115485 ns/op	58474936 B/op	    4070 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:true-4          	      10	 108524066 ns/op	  759140 B/op	    3957 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:3000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:false-4           	       2	 680593549 ns/op	58471576 B/op	    4053 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:3000,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:true-4            	      51	  22117074 ns/op	  762900 B/op	    3943 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:30,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:false-4             	       3	 509314305 ns/op	62615528 B/op	    4005 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:30,_innerRows:300000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:true-4              	      43	  28141019 ns/op	 4911220 B/op	    3900 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:330000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:false-4         	       1	1472482612 ns/op	58485000 B/op	    4072 allocs/op
BenchmarkMergeJoinExec/merge_join_(outerRows:300000,_innerRows:330000,_concurency:4,_outerJoinKeyIdx:_[0_1],_innerJoinKeyIdx:_[0_1],_NeedOuterSort:false)_InlineProj:true-4          	      10	 108598788 ns/op	  758148 B/op	    3955 allocs/op

Reminiscent

Great! LGTM

sre-bot · 2020-04-07T12:20:30Z

@lance6716 merge failed.

Reminiscent · 2020-04-07T12:24:46Z

@lance6716 Thanks for your contribution. Please resolve the conflicts. Thanks!

lance6716 · 2020-04-07T13:52:41Z

@Reminiscent conflict resolved, waiting for an approval 😄

lance6716 · 2020-04-08T04:09:04Z

ping @SunRunAway @Reminiscent

sre-bot · 2020-04-09T07:54:11Z

/run-all-tests

sre-bot · 2020-04-09T07:57:09Z

@lance6716 merge failed.

planner, executor: enable inline projection for merge join

e650691

lance6716 requested review from a team as code owners March 18, 2020 09:13

sre-bot added the contribution This PR is from a community contributor. label Mar 18, 2020

ghost requested review from francis0407, lzmhhh123, qw4990, XuHuaiyu and a team and removed request for a team March 18, 2020 09:13

lance6716 commented Mar 18, 2020

View reviewed changes

planner/core/task.go Show resolved Hide resolved

github-actions bot added the sig/execution SIG execution label Mar 18, 2020

lance6716 commented Mar 18, 2020

View reviewed changes

planner/cascades/testdata/integration_suite_out.json Outdated Show resolved Hide resolved

fix some CI

6e6056e

lance6716 added 3 commits March 18, 2020 21:16

fix more CI

017409e

fix more CI

aa4e930

untouch go.sum

60a8cd7

lance6716 changed the title ~~[WIP]planner, executor: enable inline projection for merge join~~ planner, executor: enable inline projection for merge join Mar 18, 2020

lance6716 mentioned this pull request Mar 19, 2020

Support inline projection for executors #14428

Open

9 tasks

SunRunAway reviewed Mar 19, 2020

View reviewed changes

executor/benchmark_test.go Outdated Show resolved Hide resolved

executor/benchmark_test.go Outdated Show resolved Hide resolved

executor/executor_required_rows_test.go Show resolved Hide resolved

lance6716 added 7 commits March 19, 2020 21:16

fix comment

83da3e8

Merge branch 'master' of https://github.com/pingcap/tidb into inline-…

2f1d9fb

…proj-merge-join

fix CI

5c5c5f4

fix CI

b2ab95a

fix CI

7fadc55

fix CI

9c24f1d

fix CI

256a54d

SunRunAway reviewed Mar 29, 2020

View reviewed changes

SunRunAway added the status/LGT1 Indicates that a PR has LGTM 1. label Mar 29, 2020

add benchmark

eb0c5bf

Reminiscent previously approved these changes Apr 7, 2020

View reviewed changes

Reminiscent added status/can-merge Indicates a PR has been approved by a committer. status/LGT2 Indicates that a PR has LGTM 2. and removed status/LGT1 Indicates that a PR has LGTM 1. labels Apr 7, 2020

Merge branch 'master' into inline-proj-merge-join

e40289f

lance6716 dismissed Reminiscent’s stale review via e40289f April 7, 2020 13:48

lance6716 requested review from SunRunAway and Reminiscent April 7, 2020 13:57

SunRunAway approved these changes Apr 9, 2020

View reviewed changes

SunRunAway added status/can-merge Indicates a PR has been approved by a committer. and removed status/can-merge Indicates a PR has been approved by a committer. labels Apr 9, 2020

Merge branch 'master' into inline-proj-merge-join

980bad1

SunRunAway merged commit 49ee75a into pingcap:master Apr 9, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

planner, executor: enable inline projection for merge join #15463

planner, executor: enable inline projection for merge join #15463

lance6716 commented Mar 18, 2020 •

edited

Loading

lance6716 commented Mar 18, 2020

SunRunAway left a comment

SunRunAway left a comment

lance6716 commented Mar 29, 2020

SunRunAway commented Mar 29, 2020

sre-bot commented Mar 29, 2020

lance6716 commented Mar 30, 2020

sre-bot commented Mar 30, 2020

SunRunAway commented Apr 2, 2020

sre-bot commented Apr 2, 2020

lance6716 commented Apr 2, 2020

Reminiscent left a comment

sre-bot commented Apr 7, 2020

Reminiscent commented Apr 7, 2020

lance6716 commented Apr 7, 2020 •

edited

Loading

lance6716 commented Apr 8, 2020

sre-bot commented Apr 9, 2020

sre-bot commented Apr 9, 2020

planner, executor: enable inline projection for merge join #15463

planner, executor: enable inline projection for merge join #15463

Conversation

lance6716 commented Mar 18, 2020 • edited Loading

What problem does this PR solve?

What is changed and how it works?

Related changes

Check List

Release note

benchmark based on BenchmarkMergeJoinExec

case 1

case 2

case 3

lance6716 commented Mar 18, 2020

SunRunAway left a comment

Choose a reason for hiding this comment

SunRunAway left a comment

Choose a reason for hiding this comment

lance6716 commented Mar 29, 2020

SunRunAway commented Mar 29, 2020

sre-bot commented Mar 29, 2020

Benchmark Report

lance6716 commented Mar 30, 2020

sre-bot commented Mar 30, 2020

Benchmark Report

SunRunAway commented Apr 2, 2020

sre-bot commented Apr 2, 2020

Benchmark Report

lance6716 commented Apr 2, 2020

Reminiscent left a comment

Choose a reason for hiding this comment

sre-bot commented Apr 7, 2020

Reminiscent commented Apr 7, 2020

lance6716 commented Apr 7, 2020 • edited Loading

lance6716 commented Apr 8, 2020

sre-bot commented Apr 9, 2020

sre-bot commented Apr 9, 2020

lance6716 commented Mar 18, 2020 •

edited

Loading

lance6716 commented Apr 7, 2020 •

edited

Loading