executor: support spill intermediate data for unparalleled hash agg by wshwsh12 · Pull Request #25714 · pingcap/tidb

wshwsh12 · 2021-06-23T09:40:44Z

What problem does this PR solve?

Issue Number: close #xxx

Problem Summary:

What is changed and how it works?

Proposal: Design

What's Changed:
Based PR #25820, introduce soft limit.
Support spilling intermediate date for unparalleled hashAgg.

How it Works:

When the memory usage is higher than the quota, switch the HashAgg executor to spill-mode.
When HashAgg is in spill-mode, keep the tuple in the hashMap no longer growing.
a. If the key exists in the Map, aggreagte the result.
b. If the key doesn't exist in the Map, spill the data to disk.
After all data have been processed, output the aggreagte result in the Map, clear the Map. Then read the spilling data from disk, repeat the Step1-Step3 until all data have been aggregated.

Related changes

PR to update pingcap/docs/pingcap/docs-cn:

Check List

Tests

Unit test
Integration test
Manual test (add detailed scripts or steps below)
Set AggregateConcurrency to 1 and run all correctness test in tidb repo.

Side effects

Performance regression
- Consumes more CPU
- Consumes more MEM

Release note

Support spill intermediate data for unparalleled hash agg

…-agg

wshwsh12 · 2021-06-28T06:04:16Z

Simple Benchmark

cpu: AMD Ryzen 7 3700X 8-Core Processor

Old Implement:
BenchmarkHashAggRows
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:100000,_concurrency:1,_sorted:true)
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:100000,_concurrency:1,_sorted:true)-16         	      79	  23354686 ns/op
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:1000000,_concurrency:1,_sorted:true)
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:1000000,_concurrency:1,_sorted:true)-16        	       7	 150919829 ns/op
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:10000000,_concurrency:1,_sorted:true)
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:10000000,_concurrency:1,_sorted:true)-16       	       1	1315785257 ns/op

New Implement:
BenchmarkHashAggRows
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:100000,_concurrency:1,_sorted:true)
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:100000,_concurrency:1,_sorted:true)-16         	      52	  20963917 ns/op
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:1000000,_concurrency:1,_sorted:true)
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:1000000,_concurrency:1,_sorted:true)-16        	       7	 145583208 ns/op
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:10000000,_concurrency:1,_sorted:true)
BenchmarkHashAggRows/(execType:hash,_aggFunc:sum,_ndv:1000,_hasDistinct:false,_rows:10000000,_concurrency:1,_sorted:true)-16       	       1	1327683063 ns/op

Memory usage test

Workload: tpch-sf=3g 1tidb 1tikv 1pd
Config: oom-action = "log",mem-quota-query = 1 << 30, and close copr-cache
User-veriables: tidb_hashagg_final_concurrency = 1; tidb_hashagg_partial_concurrency = 1;

Old Implement
[tidb]> desc analyze select avg(L_EXTENDEDPRICE) from lineitem group by L_PARTKEY   ,L_SUPPKEY , L_QUANTITY;
+---------------------------+-------------+----------+-----------+----------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------+------+
| id                        | estRows     | actRows  | task      | access object  | execution info                                                                                                                                                                                                                                                                             | operator info                                                                                                                                                                                   | memory   | disk |
+---------------------------+-------------+----------+-----------+----------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------+------+
| HashAgg_9                 | 14397287.20 | 16776228 | root      |                | time:4m23.7s, loops:16385                                                                                                                                                                                                                                                                  | group by:test.lineitem.l_partkey, test.lineitem.l_quantity, test.lineitem.l_suppkey, funcs:avg(Column#19, Column#20)->Column#18                                                                 | 3.14 GB  | N/A  |
| └─TableReader_10          | 14397287.20 | 17963641 | root      |                | time:3m40.2s, loops:41, cop_task: {num: 40, max: 32.5s, min: 1.55s, avg: 19.7s, p95: 31.7s, max_proc_keys: 613883, p95_proc_keys: 462536, tot_proc: 12m46.5s, tot_wait: 9.29s, rpc_num: 40, rpc_time: 13m6.7s, copr_cache: disabled}                                                       | data:HashAgg_5                                                                                                                                                                                  | 320.5 MB | N/A  |
|   └─HashAgg_5             | 14397287.20 | 17963641 | cop[tikv] |                | tikv_task:{proc max:19.4s, min:873ms, p80:18.1s, p95:19.3s, iters:17591, tasks:40}, scan_detail: {total_process_keys: 17996609, total_keys: 17996649, rocksdb: {delete_skipped_count: 0, key_skipped_count: 17996609, block: {cache_hit_count: 57170, read_count: 0, read_byte: 0 Bytes}}} | group by:test.lineitem.l_partkey, test.lineitem.l_quantity, test.lineitem.l_suppkey, funcs:count(test.lineitem.l_extendedprice)->Column#19, funcs:sum(test.lineitem.l_extendedprice)->Column#20 | N/A      | N/A  |
|     └─TableFullScan_8     | 17996609.00 | 17996609 | cop[tikv] | table:lineitem | tikv_task:{proc max:10.9s, min:511ms, p80:10.1s, p95:10.7s, iters:17591, tasks:40}                                                                                                                                                                                                         | keep order:false, stats:pseudo                                                                                                                                                                  | N/A      | N/A  |
+---------------------------+-------------+----------+-----------+----------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------+------+
4 rows in set (4 min 24.836 sec)


New Implement:
[tidb]> desc analyze select avg(L_EXTENDEDPRICE) from lineitem group by L_PARTKEY   ,L_SUPPKEY , L_QUANTITY;

+---------------------------+-------------+----------+-----------+----------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------+---------+
| id                        | estRows     | actRows  | task      | access object  | execution info                                                                                                                                                                                                                                                                             | operator info                                                                                                                                                                                   | memory   | disk    |
+---------------------------+-------------+----------+-----------+----------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------+---------+
| HashAgg_9                 | 14397287.20 | 16776228 | root      |                | time:4m8.2s, loops:16385                                                                                                                                                                                                                                                                   | group by:test.lineitem.l_partkey, test.lineitem.l_quantity, test.lineitem.l_suppkey, funcs:avg(Column#19, Column#20)->Column#18                                                                 | 819.2 MB | 3.41 GB |
| └─TableReader_10          | 14397287.20 | 17963641 | root      |                | time:52.6s, loops:41, cop_task: {num: 40, max: 30.5s, min: 1.88s, avg: 22.7s, p95: 30.1s, max_proc_keys: 613883, p95_proc_keys: 462536, tot_proc: 14m39.4s, tot_wait: 18.7s, rpc_num: 40, rpc_time: 15m8.5s, copr_cache: disabled}                                                         | data:HashAgg_5                                                                                                                                                                                  | 137.3 MB | N/A     |
|   └─HashAgg_5             | 14397287.20 | 17963641 | cop[tikv] |                | tikv_task:{proc max:19.2s, min:1.27s, p80:17.5s, p95:18.9s, iters:17591, tasks:40}, scan_detail: {total_process_keys: 17996609, total_keys: 17996649, rocksdb: {delete_skipped_count: 0, key_skipped_count: 17996609, block: {cache_hit_count: 57170, read_count: 0, read_byte: 0 Bytes}}} | group by:test.lineitem.l_partkey, test.lineitem.l_quantity, test.lineitem.l_suppkey, funcs:count(test.lineitem.l_extendedprice)->Column#19, funcs:sum(test.lineitem.l_extendedprice)->Column#20 | N/A      | N/A     |
|     └─TableFullScan_8     | 17996609.00 | 17996609 | cop[tikv] | table:lineitem | tikv_task:{proc max:10.9s, min:731ms, p80:9.91s, p95:10.7s, iters:17591, tasks:40}                                                                                                                                                                                                         | keep order:false, stats:pseudo                                                                                                                                                                  | N/A      | N/A     |
+---------------------------+-------------+----------+-----------+----------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+----------+---------+
4 rows in set (4 min 8.565 sec)

Memory usage in Grafana montior

Left is the old impelement, right is the new implement.

XuHuaiyu · 2021-07-12T08:02:04Z

+	return e.spillAction
+}
+
+const maxSpillTimes = 10


Add a comment for this

XuHuaiyu · 2021-07-12T08:03:53Z

+	listInDisk   *chunk.ListInDisk
+	lastChunkNum int
+	processIdx   int
+	spillMode    uint32
+	spillChunk   *chunk.Chunk
+	spillAction  *AggSpillDiskAction
+	childDrained bool


We need comments for these variables

XuHuaiyu · 2021-07-12T08:07:13Z

+				return err
+			}
+		}
+		if e.spillAction != nil {


What's this for?

Useless.. I remove the code now.

XuHuaiyu · 2021-07-12T08:09:51Z

 		}
+		if e.listInDisk != nil {
+			if err := e.listInDisk.Close(); err != nil {
+				return err


Don't we close the chilrenExec? This may cause leaks.

XuHuaiyu · 2021-07-12T08:12:08Z

+	e.executed, e.childDrained = false, false
+	e.listInDisk = chunk.NewListInDisk(retTypes(e.children[0]))
+	e.spillChunk = newFirstChunk(e.children[0])
+	if e.ctx.GetSessionVars().TrackAggregateMemoryUsage {


Why do we need to check this?

If tidb doesn't track aggregate executor memory usgae, should we also try to spill hashAgg when exceeded?
In addition, oom-use-tmp-storage also should be check... I add the check now. PTAL again.

XuHuaiyu · 2021-07-12T08:21:04Z

 			groupKey := string(e.groupKeyBuffer[j]) // do memory copy here, because e.groupKeyBuffer may be reused.
 			if !e.groupSet.Exist(groupKey) {
+				if atomic.LoadUint32(&e.spillMode) == 1 && e.groupSet.Count() > 0 {
+					e.spillChunk.Append(e.childResult, j, j+1)


Can we use Chunk.sel to optimize this if-block?

We can check e.groupSet.Exist(groupKey) and build the sel firstly, and then invoke e.spillChunk.Append based on the sel.

Further, if len(sel) == len(e.childResult), we can invoke e.listInDisk.Add(e.childResult) directly.

…-agg

XuHuaiyu · 2021-07-14T11:46:05Z

+
+		// spill unprocessed data when exceeded.
+		if len(sel) > 0 {
+			err = e.spillUnprocessedData(sel)


The input argument sel is useless?

e.childResult.SetSel(sel)

e.childResult.SetSel(sel) will let len(sel) == len(e.childResult) always true, and e.listInDisk.Add(e.childResult) directly. If there are only a few elements in sel, it maybe have performance issue.
I remove the logic e.listInDisk.Add(e.childResult) and always append to tmpChkForSpill, PTAL

XuHuaiyu · 2021-07-14T12:03:00Z

+	listInDisk   *chunk.ListInDisk   // listInDisk is the chunks to store row values for spilling data.
+	lastChunkNum int                 // lastChunkNum indicates the num of spilling chunk.
+	processIdx   int                 // processIdx indicates the num of processed chunk in disk.
+	spillMode    uint32              // spillMode means that no new groups are added to hash table.


isSpillModeSet?

Add an explanation for what does 0 and 1 mean

Co-authored-by: HuaiyuXu <xuhuaiyu@pingcap.com>

XuHuaiyu

LGTM

ti-chi-bot · 2021-07-15T09:05:39Z

[REVIEW NOTIFICATION]

This pull request has been approved by:

XuHuaiyu
mmyj

To complete the pull request process, please ask the reviewers in the list to review by filling /cc @reviewer in the comment.
After your PR has acquired the required number of LGTMs, you can assign this pull request to the committer in the list by filling /assign @committer in the comment to help you merge this pull request.

The full list of commands accepted by this bot can be found here.

Details

Reviewer can indicate their review by submitting an approval review.
Reviewer can cancel approval by submitting a request changes review.

XuHuaiyu · 2021-07-15T09:05:44Z

/merge

ti-chi-bot · 2021-07-15T09:05:46Z

This pull request has been accepted and is ready to merge.

Details

Commit hash: ffbcf52

ti-chi-bot · 2021-07-15T09:05:56Z

@wshwsh12: Your PR was out of date, I have automatically updated it for you.

At the same time I will also trigger all tests for you:

/run-all-tests

If the CI test fails, you just re-trigger the test that failed and the bot will merge the PR for you after the CI passes.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the ti-community-infra/tichi repository.

wshwsh12 · 2021-07-15T09:14:06Z

/run-unit-test

wshwsh12 added 3 commits June 23, 2021 11:44

support unparalleled hash agg spill

85ff675

refactor tracker

908feb1

polish code

df5f5b9

ti-chi-bot added do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. size/L Denotes a PR that changes 100-499 lines, ignoring generated files. labels Jun 23, 2021

github-actions Bot added the sig/execution SIG execution label Jun 23, 2021

add comments

73bfe27

wshwsh12 marked this pull request as ready for review June 24, 2021 03:34

wshwsh12 requested a review from a team as a code owner June 24, 2021 03:34

wshwsh12 requested review from XuHuaiyu and removed request for a team June 24, 2021 03:34

ti-chi-bot removed the do-not-merge/work-in-progress Indicates that a PR should not merge because it is a work in progress. label Jun 24, 2021

wshwsh12 added 2 commits June 28, 2021 10:15

Merge remote-tracking branch 'upstream/master' into unparalleled-hash…

cbb6b93

…-agg

disk tracker

7b87630

wshwsh12 added 2 commits June 28, 2021 14:48

fmt

39afb61

reset bInMap

bdfe307

wshwsh12 mentioned this pull request Jul 2, 2021

Support spill HashAgg to disk #25882

Closed

5 tasks

XuHuaiyu added the type/enhancement The issue or PR belongs to an enhancement. label Jul 8, 2021

mmyj approved these changes Jul 8, 2021

View reviewed changes

Comment thread executor/aggregate.go Outdated

ti-chi-bot added the status/LGT1 Indicates that a PR has LGTM 1. label Jul 8, 2021

XuHuaiyu reviewed Jul 12, 2021

View reviewed changes

ti-chi-bot added the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label Jul 12, 2021

wshwsh12 added 3 commits July 13, 2021 11:26

Merge remote-tracking branch 'upstream/master' into unparalleled-hash…

45acac8

…-agg

address comments

ca9e4fb

address comments

ac2485d

ti-chi-bot removed the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label Jul 13, 2021

optimize

c378580

wshwsh12 requested a review from XuHuaiyu July 13, 2021 07:31

add test

dfc7293

XuHuaiyu reviewed Jul 14, 2021

View reviewed changes

wshwsh12 added 2 commits July 15, 2021 10:44

address comments

000a16f

address comments

d17f694

wshwsh12 requested a review from XuHuaiyu July 15, 2021 06:21

XuHuaiyu reviewed Jul 15, 2021

View reviewed changes

address comments

6ba0908

XuHuaiyu reviewed Jul 15, 2021

View reviewed changes

Comment thread executor/aggregate.go Outdated

Update executor/aggregate.go

ffbcf52

Co-authored-by: HuaiyuXu <xuhuaiyu@pingcap.com>

XuHuaiyu approved these changes Jul 15, 2021

View reviewed changes

ti-chi-bot added status/LGT2 Indicates that a PR has LGTM 2. and removed status/LGT1 Indicates that a PR has LGTM 1. labels Jul 15, 2021

ti-chi-bot added the status/can-merge Indicates a PR has been approved by a committer. label Jul 15, 2021

Merge branch 'master' into unparalleled-hash-agg

e358d68

ti-chi-bot merged commit 2412437 into pingcap:master Jul 15, 2021

wshwsh12 deleted the unparalleled-hash-agg branch January 29, 2022 07:20

Conversation

wshwsh12 commented Jun 23, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What problem does this PR solve?

What is changed and how it works?

Related changes

Check List

Release note

Uh oh!

wshwsh12 commented Jun 28, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Simple Benchmark

Memory usage test

Memory usage in Grafana montior

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

XuHuaiyu left a comment

Choose a reason for hiding this comment

Uh oh!

ti-chi-bot commented Jul 15, 2021

Uh oh!

XuHuaiyu commented Jul 15, 2021

Uh oh!

ti-chi-bot commented Jul 15, 2021

Uh oh!

ti-chi-bot commented Jul 15, 2021

Uh oh!

wshwsh12 commented Jul 15, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

wshwsh12 commented Jun 23, 2021 •

edited

Loading

wshwsh12 commented Jun 28, 2021 •

edited

Loading