Allocate continuous row id for single INSERT statement #13648

Little-Wallace · 2019-11-21T02:54:40Z

What problem does this PR solve?

If a table is created without primary key, TiDB will assign a unique RowID to every record and store it in TiKV. This PR will make the RowID of every statement be continuous.
The continuous keys are good for TiKV to improve write performance, by skip checking write version, see TiKV #5846 for more details.

For example, we run this sql in TiDB:

CREATE TABLE t1 (
  `k1` bigint(64) DEFAULT NULL,
  `k2` bigint(64) DEFAULT NULL,
  `v1` varchar(300) DEFAULT NULL,
  `v2` int(11) NOT NULL,
   KEY `k1`(`k1`)
);
thread1: INSERT INTO t1(k1,k2,v1,v2) VALUES(1,2,'a','a'),(2,2,'a','a');
thread2: INSERT INTO t1(k1,k2,v1,v2) VALUES(2,2,'a','a'),(3,2,'a','a');

Before the PR, they may get RowID as:

thread1:  1, 3;
thread2: 2, 4;

After the PR, they will get RowID as:

thread1:  1, 2;
thread2: 3, 4;

What is changed and how it works?

I will allocate a continuous numbers every time a statement try to get row id.

Check List

Tests

Unit test

Code changes

Has interface methods change

Side effects

Increased code complexity

Related changes

No

Release note

The record in the same statement would get continuous RowID. You could use SELECT _tidb_rowid for TABLE to ensure this feature.

Signed-off-by: Little-Wallace <[email protected]>

CLAassistant · 2019-11-21T02:55:55Z

All committers have signed the CLA.

codecov · 2019-11-21T03:00:03Z

Codecov Report

Merging #13648 into master will not change coverage.
The diff coverage is n/a.

@@             Coverage Diff             @@
##             master     #13648   +/-   ##
===========================================
  Coverage   80.0389%   80.0389%           
===========================================
  Files           473        473           
  Lines        116507     116507           
===========================================
  Hits          93251      93251           
  Misses        15926      15926           
  Partials       7330       7330

Signed-off-by: Little-Wallace <[email protected]>

zhangjinpeng87 · 2019-11-21T08:56:29Z

Please describe the benefits for continuous row id.

zhangjinpeng87 · 2019-11-21T09:00:20Z

table/tables/tables.go

+					return 0, err
+				}
+			}
+			stmtCtx.BaseRowID += 1


Should increase BaseRowID after L468?

zhangjinpeng87 · 2019-11-21T09:02:26Z

table/tables/tables.go

+
+// AllocHandle implements table.Table AllocHandle interface.
+func (t *tableCommon) AllocHandleIDs(ctx sessionctx.Context, n uint64) (int64, int64, error) {
+	base, rowID, err := t.Allocator(ctx).Alloc(t.tableID, n)


s/rowID/maxID

Signed-off-by: Little-Wallace <[email protected]>

zhangjinpeng87

LGTM

Little-Wallace · 2019-11-21T10:10:13Z

This PR could reduce more than 15% CPU of TiKV server when insert many keys.

The query insert 1 is sysbench prepare.
The query insert 2 is batch_insert of benchbot. See it at (https://github.com/pingcap/productivity-tools/blob/master/benchmark/batch_insert/batch_insert.lua#L75)

winoros · 2019-11-21T10:31:05Z

table/tables/tables.go

+		rows := stmtCtx.RecordRows()
+		if rows > 1 {
+			if stmtCtx.BaseRowID >= stmtCtx.MaxRowID {
+				stmtCtx.BaseRowID, stmtCtx.MaxRowID, err = t.AllocHandleIDs(ctx, rows)


The MaxRowID is not used actually?

sre-bot · 2019-11-21T11:35:31Z

Benchmark Report

Run Sysbench Performance Test on VMs

@@                               Benchmark Diff                               @@
================================================================================
--- tidb: ce87b6c6713d551b3700ea9732cd0955a3819952
+++ tidb: fed318f81094296d48db98ed86d2053b2670414a
tikv: ae6038ea22271b23f2290e61e45838281ede4e20
pd: c71ec9500e49c933c198bba1f901a5b42a72de86
================================================================================
oltp_update_non_index:
    * QPS: 4723.07 ± 0.06% (std=1.99) delta: 0.20% (p=0.008)
    * Latency p50: 27.10 ± 0.06% (std=0.01) delta: -0.26%
    * Latency p99: 41.48 ± 0.90% (std=0.38) delta: -4.84%
            
oltp_insert:
    * QPS: 4775.71 ± 0.03% (std=0.93) delta: 0.21% (p=0.184)
    * Latency p50: 26.80 ± 0.02% (std=0.00) delta: -0.21%
    * Latency p99: 47.70 ± 2.23% (std=0.71) delta: -3.55%
            
oltp_read_write:
    * QPS: 15658.36 ± 0.69% (std=66.24) delta: 0.81% (p=0.039)
    * Latency p50: 163.84 ± 0.65% (std=0.66) delta: -0.78%
    * Latency p99: 303.45 ± 3.62% (std=8.64) delta: -1.34%
            
oltp_update_index:
    * QPS: 4253.34 ± 0.40% (std=12.51) delta: -0.09% (p=0.393)
    * Latency p50: 30.09 ± 0.40% (std=0.09) delta: 0.09%
    * Latency p99: 53.21 ± 2.39% (std=0.90) delta: -3.39%
            
oltp_point_select:
    * QPS: 40310.62 ± 0.22% (std=67.89) delta: 0.19% (p=0.527)
    * Latency p50: 3.17 ± 0.16% (std=0.01) delta: -0.16%
    * Latency p99: 9.73 ± 0.00% (std=0.00) delta: 1.78%

sre-bot · 2019-11-22T08:19:06Z

/run-all-tests

…ap#13648)" This reverts commit 71e19a7.

tiancaiamao · 2020-02-07T10:01:22Z

executor/insert_test.go

+				}
+				tk.MustExec(sql)
+				q := "select _tidb_rowid from t1 where a=" + k
+				fmt.Printf("query: %v\n", q)


@Little-Wallace

tiancaiamao · 2020-03-09T17:24:53Z

table/tables/tables.go

-		if err != nil {
-			return 0, err
+		stmtCtx := ctx.GetSessionVars().StmtCtx
+		rows := stmtCtx.RecordRows()


This number is wrong...
It's not the row counts in this statement, it's a batch of chunk rows and accumulative.

Little-Wallace added 3 commits November 20, 2019 13:33

allocate range row id for insert

20310a0

Signed-off-by: Little-Wallace <[email protected]>

small fix

aedd015

Signed-off-by: Little-Wallace <[email protected]>

small fix

7da26e5

Signed-off-by: Little-Wallace <[email protected]>

Little-Wallace added status/WIP sig/execution SIG execution labels Nov 21, 2019

Little-Wallace changed the title ~~Batch row~~ Allocate continuous row id for INSERT statement Nov 21, 2019

winoros requested review from lysu, coocood and jackysp November 21, 2019 04:34

Little-Wallace added 3 commits November 21, 2019 15:30

add unit test

4d7732c

Signed-off-by: Little-Wallace <[email protected]>

Merge branch 'master' into batch_row_id

a51c01d

small fix

45d11ae

Signed-off-by: Little-Wallace <[email protected]>

Little-Wallace removed the status/WIP label Nov 21, 2019

zhangjinpeng87 reviewed Nov 21, 2019

View reviewed changes

table/tables/tables.go

return 0, err

}

}

stmtCtx.BaseRowID += 1

Copy link

Contributor

zhangjinpeng87 Nov 21, 2019

Choose a reason for hiding this comment

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

Should increase BaseRowID after L468?

zhangjinpeng87 reviewed Nov 21, 2019

View reviewed changes

Little-Wallace added 2 commits November 21, 2019 17:21

fix variable name

4025063

Signed-off-by: Little-Wallace <[email protected]>

small fix

a6d9262

Signed-off-by: Little-Wallace <[email protected]>

zhangjinpeng87 approved these changes Nov 21, 2019

View reviewed changes

Merge branch 'master' into batch_row_id

fed318f

zhangjinpeng87 changed the title ~~Allocate continuous row id for INSERT statement~~ Allocate continuous row id for single INSERT statement Nov 21, 2019

winoros requested changes Nov 21, 2019

View reviewed changes

Merge branch 'master' into batch_row_id

1cff8c1

winoros approved these changes Nov 22, 2019

View reviewed changes

Merge branch 'master' into batch_row_id

f2844a1

Little-Wallace added the status/can-merge Indicates a PR has been approved by a committer. label Nov 22, 2019

sre-bot merged commit 71e19a7 into pingcap:master Nov 22, 2019

cfzjywxk added a commit to cfzjywxk/tidb that referenced this pull request Dec 8, 2019

Revert "Allocate continuous row id for single INSERT statement (pingc…

ab77b7e

…ap#13648)" This reverts commit 71e19a7.

cfzjywxk mentioned this pull request Dec 8, 2019

executor: add log for commit work #13965

Merged

XiaTianliang pushed a commit to XiaTianliang/tidb that referenced this pull request Dec 21, 2019

Allocate continuous row id for single INSERT statement (pingcap#13648)

ef1ef03

tiancaiamao reviewed Feb 7, 2020

View reviewed changes

tiancaiamao reviewed Mar 9, 2020

View reviewed changes

tiancaiamao mentioned this pull request Mar 10, 2020

*: fix autoid allocation to avoid quickly exhaustion #15261

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allocate continuous row id for single INSERT statement #13648

Allocate continuous row id for single INSERT statement #13648

Little-Wallace commented Nov 21, 2019 •

edited

Loading

CLAassistant commented Nov 21, 2019 •

edited

Loading

codecov bot commented Nov 21, 2019 •

edited

Loading

zhangjinpeng87 commented Nov 21, 2019

zhangjinpeng87 Nov 21, 2019

zhangjinpeng87 Nov 21, 2019

zhangjinpeng87 left a comment

Little-Wallace commented Nov 21, 2019

winoros Nov 21, 2019

sre-bot commented Nov 21, 2019

sre-bot commented Nov 22, 2019

tiancaiamao Feb 7, 2020

zhangjinpeng87 Feb 7, 2020

tiancaiamao Mar 9, 2020

Allocate continuous row id for single INSERT statement #13648

Allocate continuous row id for single INSERT statement #13648

Conversation

Little-Wallace commented Nov 21, 2019 • edited Loading

What problem does this PR solve?

What is changed and how it works?

Check List

CLAassistant commented Nov 21, 2019 • edited Loading

codecov bot commented Nov 21, 2019 • edited Loading

Codecov Report

zhangjinpeng87 commented Nov 21, 2019

zhangjinpeng87 Nov 21, 2019

Choose a reason for hiding this comment

zhangjinpeng87 Nov 21, 2019

Choose a reason for hiding this comment

zhangjinpeng87 left a comment

Choose a reason for hiding this comment

Little-Wallace commented Nov 21, 2019

winoros Nov 21, 2019

Choose a reason for hiding this comment

sre-bot commented Nov 21, 2019

Benchmark Report

sre-bot commented Nov 22, 2019

tiancaiamao Feb 7, 2020

Choose a reason for hiding this comment

zhangjinpeng87 Feb 7, 2020

Choose a reason for hiding this comment

tiancaiamao Mar 9, 2020

Choose a reason for hiding this comment

Little-Wallace commented Nov 21, 2019 •

edited

Loading

CLAassistant commented Nov 21, 2019 •

edited

Loading

codecov bot commented Nov 21, 2019 •

edited

Loading