pingcap / pingcap/tiflash

align null flags of union when making logical plans

Open
#3,331 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

component/mpp type/enhancement
Dominant language
C++
Stars
1k
Forks
423
Avg merge
1d 15h
Merged PRs (30d)
24

Description

Use test;
create table t (id int, d double, nd double not null);
Alter table t set tiflash replica 1;
insert into t values (0,0,0),(1,1,1),(2,null,2);
create table tt like t;
alter table tt set tiflash replica 1;
insert into tt  select * from t;
insert into tt  select * from t;
insert into tt  select * from t;
insert into tt  select * from t;

mysql> desc Select IDD from tt join ( (select 127 as IDD from t) union all (select 1 as IDD from t) ) u on tt.id = u.IDD;
+--------------------------------------------+---------+--------------+---------------+----------------------------------------------+
| id                                         | estRows | task         | access object | operator info                                |
+--------------------------------------------+---------+--------------+---------------+----------------------------------------------+
| Projection_13                              | 48.00   | root         |               | Column#15                                    |
| └─TableReader_27                           | 48.00   | root         |               | data:ExchangeSender_26                       |
|   └─ExchangeSender_26                      | 48.00   | cop[tiflash] |               | ExchangeType: PassThrough                    |
|     └─HashJoin_14                          | 48.00   | cop[tiflash] |               | inner join, equal:[eq(fzh.tt.id, Column#15)] |
|       ├─ExchangeReceiver_25(Build)         | 12.00   | cop[tiflash] |               |                                              |
|       │ └─ExchangeSender_24                | 12.00   | cop[tiflash] |               | ExchangeType: Broadcast                      |
|       │   └─Union_19                       | 12.00   | cop[tiflash] |               |                                              |
|       │     ├─Projection_48                | 6.00    | cop[tiflash] |               | cast(Column#15, bigint(3) BINARY)->Column#15 |
|       │     │ └─Projection_20              | 6.00    | cop[tiflash] |               | 127->Column#15                               |
|       │     │   └─TableFullScan_21         | 6.00    | cop[tiflash] | table:t       | keep order:false, stats:pseudo               |
|       │     └─Projection_22                | 6.00    | cop[tiflash] |               | 1->Column#15                                 |
|       │       └─TableFullScan_23           | 6.00    | cop[tiflash] | table:t       | keep order:false, stats:pseudo               |
|       └─Selection_18(Probe)                | 23.98   | cop[tiflash] |               | not(isnull(fzh.tt.id))                       |
|         └─TableFullScan_17                 | 24.00   | cop[tiflash] | table:tt      | keep order:false, stats:pseudo               |
+--------------------------------------------+---------+--------------+---------------+----------------------------------------------+
14 rows in set (0.00 sec)
|       │     ├─Projection_48                | 6.00    | cop[tiflash] |               | cast(Column#15, bigint(3) BINARY)->Column#15 |

the projection_48 is forced adding after making logical plan, actually at making MPP plan. In fact, we should enfoce the logic to union data types as well as null flags duing making logical plan. In details, when making logical plan, the 127 branch is the same as 1 branch, only one projection, but MPP considers null flags, then forcing a projection later.
why only 127 branch needs a proj? where added?

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No source files, tests, or entry points are named. Start by reproducing the SQL plan in the issue and compare the logical and MPP plans, then trace where the extra Projection_48 is introduced. Done means the union's data types and null flags are aligned during logical-plan construction without the later forced projection.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, sql
Domain
backend, databases
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.