*: Support required rows rows for arrow decode format.

[wshwsh12]

What problem does this PR solve?

When using arrow format to decode data, it appends all data into one chunk.
It may cause that a chunk save too many things.
In this pr, support required rows for arrow decode format.

What is changed and how it works?

Add a struct ArrowDecoder to deal with the detail when append required rows to the returned chunk.

1. Use the data bytes slice to build a chunk in the struct. (Memory reuse, really fast)
2. Append required rows data from the struct to the returned chunk, and record the information that how many data has been used.

Check List

Tests

• Unit test
• Integration test

Code changes

• Has exported function/method change
• Has persistent data change

Side effects

• Possible performance regression
• Increased code complexity

Related changes

Release note

Benchmark

BenchmarkSelectResponseChunk_SmallResponse-8        3006            397800 ns/op
BenchmarkSelectResponseChunk_SmallResponse_master-8   	    2949	    414161 ns/op
BenchmarkSelectResponseChunk_BigResponse-8   	     868	   1360729 ns/op
BenchmarkSelectResponseChunk_BigResponse_master-8   	     872	   1370129 ns/op

name                               old time/op  new time/op  delta
SelectResponseChunk_BigResponse-8  1.40ms ± 1%  1.41ms ± 3%   ~     (p=0.690 n=5+5)
SelectResponseChunk_SmallResponse-8   386µs ± 3%   405µs ± 1%  +4.94%  (p=0.008 n=5+5)

TPCH 50G

Version:
tidb: 5340e177b6be541f2981a9e29b7bf7a3625f7826(master) 2019-11-03 06:46:04
tikv: 086056fabd7f4763d111856854ed18f6ec9bd9da(master) 2019-11-03 02:28:24
pd: ba01076ee005b5545d679dd903f2674416176f09(master) 2019-11-01 09:12:47

Compare:
tidb: https://github.com/pingcap/tidb/compare/11cd41d...5340e17
tikv: https://github.com/tikv/tikv/compare/292ce44...086056f
pd: https://github.com/pingcap/pd/compare/ba01076...ba01076

01.sql elapsed_ms: 67776.55 ± 0.2802% (std=144.05) delta: 0.24%
02.sql elapsed_ms: 21576.88 ± 0.4347% (std=58.89) delta: -0.02%
03.sql elapsed_ms: 80073.02 ± 1.6410% (std=1017.39) delta: 0.05%
04.sql elapsed_ms: 38953.94 ± 1.5387% (std=474.05) delta: 0.57%
06.sql elapsed_ms: 44458.88 ± 0.2088% (std=63.58) delta: -0.82%
07.sql elapsed_ms: 61915.68 ± 0.4960% (std=185.78) delta: -1.17%
08.sql elapsed_ms: 65450.44 ± 0.9757% (std=450.77) delta: 1.53%
09.sql elapsed_ms: 136934.56 ± 2.1912% (std=2104.63) delta: -2.93%
10.sql elapsed_ms: 55088.41 ± 0.4401% (std=164.90) delta: -0.42%
11.sql elapsed_ms: 12339.98 ± 1.6768% (std=127.05) delta: -0.01%
12.sql elapsed_ms: 52506.32 ± 2.2628% (std=733.97) delta: -0.82%
13.sql elapsed_ms: 49567.83 ± 1.4129% (std=462.97) delta: 0.80%
14.sql elapsed_ms: 48402.44 ± 1.6835% (std=496.35) delta: -0.85%
15.sql elapsed_ms: 70156.01 ± 0.0742% (std=37.20) delta: -1.41%
16.sql elapsed_ms: 28360.51 ± 1.2965% (std=308.93) delta: -1.22%
17.sql elapsed_ms: 121267.60 ± 0.8500% (std=841.52) delta: -1.90%
18.sql elapsed_ms: 232786.07 ± 2.4822% (std=4084.29) delta: 0.45%
19.sql elapsed_ms: 54810.48 ± 0.3490% (std=123.62) delta: -0.82%
20.sql elapsed_ms: 53344.90 ± 1.2403% (std=450.87) delta: -0.98%
21.sql elapsed_ms: 96147.97 ± 0.3220% (std=263.72) delta: -0.68%
22.sql elapsed_ms: 21674.11 ± 2.1520% (std=308.77) delta: -3.76%

[codecov]

Codecov Report

Merging #12613 into master will decrease coverage by 0.0044%.
The diff coverage is 93.9024%.

@@               Coverage Diff                @@
##             master     #12613        +/-   ##
================================================
- Coverage   80.0972%   80.0927%   -0.0045%     
================================================
  Files           468        468                
  Lines        110814     110794        -20     
================================================
- Hits          88759      88738        -21     
- Misses        15200      15201         +1     
  Partials       6855       6855

[XuHuaiyu]

1. Does TiKV split the data into multiple byte slices when returning to TiDB?
2. Is there any test result that shows how much performance regression or improvement after this commit?

[wshwsh12]

/run-all-tests

[zyxbest]

/run-mybatis-test

[SunRunAway]

Consider rearrange our benchmark targeting SelectResponse, writing two benchmark functions BenchmarkSelectResponseArror_BigResponse and BenchmarkSelectResponseArror_SmallResponse is ok.

[SunRunAway]

LGTM

[XuHuaiyu]

LGTM

[sre-bot]

Your auto merge job has been accepted, waiting for 12669

[sre-bot]

/run-all-tests