risinglight
risinglight copied to clipboard
risinglightdb
executor: better explain for physical plans
Currently, RisingLight's explain interface is not really "explainable". There are some executors that...
- Use Rust's default
std::fmt::Debugwhen explaining - Doesn't print out all information
- Output too long line
- ...
We should fix those plans and make the explain UI more pretty.
A (previous) bad example is TPC-H Q1:
-
PhysicalProjection's output is too long. Maybe we should insert new lines forexprs? -
PhysicalHashAggdoesn't print out its aggregation functions.
> explain select
l_returnflag,
l_linestatus,
sum(l_quantity) as sum_qty,
sum(l_extendedprice) as sum_base_price,
sum(l_extendedprice * (1 - l_discount)) as sum_disc_price,
sum(l_extendedprice * (1 - l_discount) * (1 + l_tax)) as sum_charge,
avg(l_quantity) as avg_qty,
avg(l_extendedprice) as avg_price,
avg(l_discount) as avg_disc,
count(*) as count_order
from
lineitem
where
l_shipdate <= date '1998-12-01' - interval '71' day
group by
l_returnflag,
l_linestatus
order by
l_returnflag,
l_linestatus;
+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| PhysicalOrder: [InputRef #0 (asc), InputRef #1 (asc)] |
| PhysicalProjection: exprs [InputRef #0, InputRef #1, InputRef #2 (alias to sum_qty), InputRef #3 (alias to sum_base_price), InputRef #4 (alias to sum_disc_price), InputRef #5 (alias to sum_charge), Divide(InputRef #6, InputRef #7 as Decimal(Some(15), Some(2))) (alias to avg_qty), Divide(InputRef #8, InputRef #9 as Decimal(Some(15), Some(2))) (alias to avg_price), Divide(InputRef #10, InputRef #11 as Decimal(Some(15), Some(2))) (alias to avg_disc), InputRef #12 (alias to count_order)] |
| PhysicalHashAgg: 11 agg calls |
| PhysicalTableScan: table #7, columns [10, 8, 9, 4, 5, 6, 7, 0], with_row_handler: false, is_sorted: false, expr: LtEq(InputRef #0, Date(Date(10490)) (const)) |
| |
+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
This is relatively an easy task to do. If you have interest in improving the explain output, feel free to send PRs!
What we need to do is simply change the Display implementation:
impl fmt::Display for PhysicalFilter {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
writeln!(f, "PhysicalFilter: expr {:?}", self.logical().expr())
}
}
... to match the format of other queries. Like:
PhysicalOrder: [InputRef #0 (asc), InputRef #1 (asc)]
PhysicalProjection:
InputRef #0
InputRef #1
InputRef #2 (alias to sum_qty)
InputRef #3 (alias to sum_base_price)
InputRef #4 (alias to sum_disc_price)
InputRef #5 (alias to sum_charge)
(InputRef #6 / InputRef #7) (alias to avg_qty)
(InputRef #8 / InputRef #9) (alias to avg_price)
(InputRef #10 / InputRef #11) (alias to avg_disc)
InputRef #12 (alias to count_order)
PhysicalHashAgg:
sum(InputRef #3) -> NUMERIC(15,2)
sum(InputRef #4) -> NUMERIC(15,2)
sum((InputRef #4 * (1 - InputRef #5))) -> NUMERIC(15,2) (null)
sum(((InputRef #4 * (1 - InputRef #5)) * (1 + InputRef #6))) -> NUMERIC(15,2) (null)
sum(InputRef #3) -> NUMERIC(15,2)
count(InputRef #3) -> INT
sum(InputRef #4) -> NUMERIC(15,2)
count(InputRef #4) -> INT
sum(InputRef #5) -> NUMERIC(15,2)
count(InputRef #5) -> INT
count(InputRef #0) -> INT
PhysicalTableScan:
table #7,
columns [10, 8, 9, 4, 5, 6, 7],
with_row_handler: true,
is_sorted: false,
expr: LtEq(InputRef #0, Date(Date(10490)) (const))
- [x] HashAgg, Projection and TableScan https://github.com/risinglightdb/risinglight/pull/469
- [ ] Drop
- [ ] Filter
- [ ] Order
- [x] HashJoin
- [ ] NestedLoopJoin
- [ ] TopN
- [ ] Values
Currently, the NLJ will output:
> create table t1(v1 int not null);
created
in 22ms 578us 583ns
> create table t2(v1 int not null);
created
in 22ms 385us 875ns
> explain select * from t1, t2 where true;
PhysicalProjection:
InputRef #0
InputRef #1
PhysicalNestedLoopJoin: op Inner, predicate: Bool(true) (const)
PhysicalTableScan:
table #8,
columns [0],
with_row_handler: false,
is_sorted: false,
expr: None
PhysicalTableScan:
table #9,
columns [0],
with_row_handler: false,
is_sorted: false,
expr: None
while order by will output:
> explain select * from t1 order by v1;
PhysicalOrder: [InputRef #0 (asc)]
PhysicalProjection:
InputRef #0
PhysicalTableScan:
table #8,
columns [0],
with_row_handler: false,
is_sorted: false,
expr: None
We should change OrderBy and NLJ to the multi-line format.
cc @WindowsXp-Beta will take over this, thanks!
Hi @skyzh , I am relatively new to Rust and RisingLight, but I am very interested in contributing to the project!
Currently, I think this issue has been resolved for
- OrderBy
- NLJ
- HashAgg, Projection, TableScan
- HashJoin
How about the rest like
- Drop
- Filter
- TopN
- Values
Can I kindly check if we have some target (improved) output for their displays?
Many thanks!
I think what you've mentioned can be improved to show as much information as possible, so you can have a try if you want!
I think what you've mentioned can be improved to show as much information as possible, so you can have a try if you want!
Sure! I will have a closer look at the related methods for Drop, Filter, TopN, Values and give it a try:) Do allow me to take a few days to understand the context and codebase a bit more in-depth.
Thanks again!
Hi, @skyzh @wangrunji0408 let me try drop. ps: I may need a little time, I'm a beginner in rust
Hi, @skyzh let me try drop
Apologies that I didn't have the bandwidth to complete this. Pls feel free to assign to another person. Thanks.
@caicancai Just go ahead!
@litone01 Never mind. You don't need to apologize. This is normal in open source and we are human beings.
@caicancai Just go ahead!
@litone01 Never mind. You don't need to apologize. This is normal in open source and we are human beings.
> explain select * from t1, t2 where true;
+----------------------------------------------------------------------+
| Join { type: inner, cost: 0, rows: 0 } |
| ├── Scan { table: t1, list: [ v1 ], filter: null, cost: 0, rows: 0 } |
| └── Scan { table: t2, list: [ v1 ], filter: null, cost: 0, rows: 0 } |
+----------------------------------------------------------------------+
Hello, I found that this part of the overall code seems to be refactured, this task seems to be no longer applicable to the current code structure, the presentation of the entire physical plan is not consistent with the issue description