4paradigm / 4paradigm/OpenMLDB
When ts=0, the calculation of history window is wrong
- Dominant language
- C++
- Stars
- 1.7k
- Forks
- 331
- Avg merge
- 12d 12h
- Merged PRs (30d)
- 1
Description
**Bug Description**
When ts=0, the calculation of history window is wrong.
+ case
```yaml
- id: 32
desc: rows_range-纯历史窗口-current_row-ts=0
inputs:
- columns: [ "c1 string","c3 int","c4 bigint","c5 float","c6 double","c7 timestamp","c8 date" ]
indexs: [ "index1:c1:c7" ]
rows:
- [ "aa",20,30,1.1,2.1,0,"2020-05-01" ]
- [ "aa",21,31,1.2,2.2,0,"2020-05-02" ]
- [ "aa",22,32,1.3,2.3,1000,"2020-05-03" ]
- [ "aa",23,33,1.4,2.4,2000,"2020-05-04" ]
- [ "bb",24,34,1.5,2.5,1590738994000,"2020-05-05" ]
sql: |
SELECT c1, c3, sum(c4) OVER w1 as w1_c4_sum FROM {0} WINDOW w1 AS (PARTITION BY {0}.c1 ORDER BY {0}.c7 ROWS_RANGE BETWEEN 2s PRECEDING AND 1s PRECEDING EXCLUDE CURRENT_ROW);
expect:
order: c3
columns: [ "c1 string","c3 int","w1_c4_sum bigint" ]
rows:
- [ "aa",20,null ]
- [ "aa",21,null ]
- [ "aa",22,61 ]
- [ "aa",23,93 ]
- [ "bb",24,null ]
```
**Expected Behavior**

**Relation Case**
test_current_row.yaml id:32
**Steps to Reproduce**
1.
2.
3.
4.
Contributor guide
Research direction
The bug is in the calculation of the history window when timestamp (ts) is 0, as shown in the YAML test case. Look at the window function logic in the codebase, likely in the SQL engine or window computation module. The test case 'test_current_row.yaml id:32' is the failing test; run it to reproduce the issue. Investigate how timestamps of zero are handled in the rows_range window frame, particularly for the '2s PRECEDING AND 1s PRECEDING EXCLUDE CURRENT_ROW' clause.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- sql
- Domain
- databases, machine-learning
- Issue type
- Bug
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 45/100